AI-generated summaries
Today's ML research,
without the noise.
Daily summaries of the latest machine learning papers from arXiv, processed every 8 hours.
41
Papers today
8h
Update frequency
7
Days of history
Correlation-Guided Fast Machine Unlearning via Hessian Analysis
Efficient ML
Theory
- Introduces a computationally efficient unlearning framework using Pearson correlation.
- Derives a closed-form parameter update rule that eliminates costly Hessian computations.
- Establishes theoretical guarantees and error bounds for the proposed method.
- Demonstrates significant speedup and improved accuracy compared to existing unlearning techniques.
Read more
Correlation-Guided Fast Machine Unlearning via Hessian Analysis
Summary
This paper addresses the pressing need for efficient machine unlearning mechanisms in the context of network security and privacy regulations like GDPR. The authors propose a novel framework that leverages Pearson correlation to identify correlated data points in the training set, allowing for a computationally efficient unlearning process. Traditional unlearning methods often rely on expensive Hessian-inverse-vector computations, which become prohibitive when multiple data points need to be removed. The proposed method introduces a closed-form parameter update rule derived from the Sherman-Morrison formula, significantly reducing computational overhead while maintaining model performance. The authors establish a theoretical relationship between high Pearson correlation and gradient proportionality for quadratic loss, ensuring that their method is robust and effective. Comprehensive evaluations across seven diverse datasets demonstrate that their approach achieves an 82× speedup over standard methods while improving accuracy. The results indicate superior forgetting effectiveness, making this method particularly suitable for real-world applications in security systems where rapid response to data removal requests is essential.
Methodology
The authors utilize Pearson correlation to identify correlated data points and derive a closed-form parameter update rule using the Sherman-Morrison formula. This approach avoids the need for repeated Hessian-inverse-vector computations, thus enhancing computational efficiency. The method is evaluated across multiple datasets and architectures to assess its effectiveness in forgetting specific training data.
Results
The proposed framework achieves an 82× speedup in wall-clock time compared to standard influence function unlearning methods while preserving model utility with a 10−2 improvement in accuracy. The evaluation shows superior forgetting effectiveness, with membership inference attack success rates of 0.660 and tug-of-war scores of 0.950 across seven diverse datasets.
Implications
This research has significant implications for the development of ethical AI systems, particularly in sectors requiring compliance with privacy regulations and the ability to efficiently manage data removal requests. The proposed method can enhance the security and privacy of machine learning models in real-world applications, such as intrusion detection systems and spam filters.
Attention Quantization for Tabular Foundation Models
Efficient ML
- Focus on attention calculation quantization rather than weight quantization for tabular models.
- Development of a quantization strategy for queries, keys, and values to FP8 format.
- Achieved up to 1.7x speedup in attention calculations with minimal accuracy loss.
- Importance of aligning quantization errors between training and testing datasets.
Read more
Attention Quantization for Tabular Foundation Models
Summary
This paper addresses the optimization of inference performance in tabular foundation models, which have gained traction similar to transformer-based large language models (LLMs). The authors propose a novel quantization strategy focused on the attention calculation rather than weight or KV cache quantization, which are more common in LLMs. They introduce a method to quantize queries, keys, and values to FP8 format, leveraging explicit FP8 matrix multiplication instructions to enhance the speed of attention calculations. A critical finding is that aligning the quantization errors between training and testing rows is essential to maintain accuracy. The proposed Triton kernel demonstrates a speedup of up to 1.7x compared to standard 16-bit kernels, with no significant accuracy loss observed across various datasets, including TabPFN-v3 and TabICLv2. This work represents the first exploration of FP8 attention in tabular foundation models, suggesting that such quantization can be effectively integrated into existing models post-training.
Methodology
The authors implemented a quantization strategy for the attention mechanism in tabular foundation models, specifically targeting the matrix multiplications involved in the attention calculation. They utilized the e4m3fn FP8 format for quantization and dynamically computed scales for each tensor to minimize quantization error. The approach was evaluated using a Triton kernel designed for efficient FP8 matrix multiplications, focusing on the in-context learning (ICL) stage of the transformer architecture.
Results
The proposed method achieved a speedup of up to 1.7x over traditional 16-bit kernels, particularly when the number of training rows exceeded 8192. The accuracy remained stable across different datasets, with no relevant loss observed in performance metrics on TabArena and BeyondArena benchmarks. The results indicate that the quantization strategy is effective for maintaining model performance while significantly enhancing computational efficiency.
Implications
This research has significant implications for the deployment of tabular foundation models in resource-constrained environments, where inference speed and efficiency are critical. The findings suggest that FP8 quantization can be a viable approach for improving the performance of machine learning models on modern hardware, potentially leading to broader adoption of tabular models in practical applications.
ProactiveBench: Can Streaming Video Models Really Interact Like Humans?
Multimodal
- ProactiveBench evaluates streaming video models at one-second intervals without explicit response cues.
- The framework includes six subtasks that assess different aspects of response timing and trigger clarity.
- Premature responses are identified as the predominant error in the evaluated systems.
- The study reveals a substantial gap in the temporal decision-making required for human-like interaction.
Read more
ProactiveBench: Can Streaming Video Models Really Interact Like Humans?
Summary
The paper introduces ProactiveBench, a novel evaluation framework for streaming video understanding models that emphasizes proactive interaction. Unlike traditional benchmarks that assess models at predetermined timestamps, ProactiveBench evaluates models continuously at one-second intervals without explicit response cues. This approach allows for a more realistic assessment of a model's ability to monitor ongoing requests and respond appropriately based on temporal context. The framework includes six subtasks that vary in trigger ambiguity and timing tolerance, enabling a comprehensive evaluation of response behaviors. Key findings indicate that premature responses are more common than missed responses across multiple evaluated systems, highlighting a significant gap in the temporal decision-making capabilities of these models compared to human-like interaction. The study emphasizes the need for improved response timing and decision-making in streaming video models to enhance their interactive capabilities.
Methodology
The authors developed ProactiveBench to evaluate models on their ability to respond to standing requests in streaming video contexts. The evaluation protocol involves continuous monitoring of video streams at one-second intervals, where models must decide whether to respond or remain silent based on the evidence available. The six subtasks are designed to vary in terms of response timing tolerance and semantic clarity, allowing for a nuanced assessment of model performance.
Results
The evaluation of six different systems using ProactiveBench revealed that premature responses were more frequent than missed responses in four out of six cases. This finding indicates that while models can respond, they often do so too early, lacking the necessary temporal decision-making to mimic human interaction effectively. The results also highlighted the importance of paired scoring to accurately assess model performance.
Implications
The findings suggest that current streaming video models require significant improvements in their temporal decision-making capabilities to achieve human-like interaction. ProactiveBench provides a framework for future research to develop models that can better manage response timing and silence, potentially leading to more effective applications in real-time video processing and interactive systems.
Performance, Efficiency and Collapse -- Advantages and Challenges in Offline Post-training of Code LLMs
Large Language Models
Reinforcement Learning
Efficient ML
- Offline RL can significantly improve the performance of code-generating LLMs without online sampling.
- The study reveals that model performance is sensitive to learning rates and training epochs.
- Prolonged training can lead to model collapse, necessitating early stopping strategies.
- Logit variance is identified as a key source of instability in offline RL training.
Read more
Performance, Efficiency and Collapse -- Advantages and Challenges in Offline Post-training of Code LLMs
Summary
This paper investigates the post-training phase of code-generating large language models (LLMs) using offline reinforcement learning (RL). The authors highlight the computational challenges associated with traditional online post-training, which requires generating new code samples and incurs significant GPU-CPU communication overhead. They propose a novel approach that leverages existing datasets for offline RL, demonstrating that substantial improvements in zero-shot code generation performance can be achieved without the need for online sampling. The study encompasses various model families and sizes, revealing that offline RL can enhance performance across models ranging from 0.5B to 7B parameters, although the degree of improvement varies. The authors also explore the sensitivity of model performance to learning rates and training epochs, noting that while multiple epochs can enhance performance, excessive training may lead to model collapse. They identify logit variance as a primary source of instability in offline RL training and propose diagnostics for early stopping to mitigate this issue. This work represents a significant contribution to the understanding of offline RL in the context of code LLMs, paving the way for more efficient post-training methodologies.
Methodology
The authors conducted experiments using offline RL on various model families (Qwen Coder, DeepSeek Coder, CodeLlama) and sizes, utilizing the CodeNet dataset. They analyzed the impact of learning rates and training epochs on model performance and investigated the causes of instability during offline RL training.
Results
The results demonstrated that offline RL can lead to significant performance improvements in code generation tasks across different model families and sizes. However, the performance gains were sensitive to the choice of learning rates and the number of training epochs, with excessive training leading to model collapse. Early stopping based on logit variance was proposed as a solution to mitigate this issue.
Implications
The findings suggest that offline RL can be a viable alternative to traditional online post-training methods, potentially reducing computational costs and improving efficiency in the development of code-generating LLMs. This approach may also facilitate the exploration of additional feedback types beyond functional correctness in future research.
Theoretical Guarantees for One-Shot Magnitude Pruning and Compute-Adaptive Early Exit
Theory
Efficient ML
- Establishes a concentration theorem for one-shot magnitude pruning in single neurons.
- Introduces the conditional perceptron model for adaptive early exit with proven generalization error decay.
- Characterizes the accumulation of pruning distortions in deep networks and derives compute-accuracy tradeoffs.
- Provides numerical simulations that support theoretical predictions regarding compute reduction mechanisms.
Read more
Theoretical Guarantees for One-Shot Magnitude Pruning and Compute-Adaptive Early Exit
Summary
This paper investigates compute reduction in neural networks through a unified perspective on partial versus full computation, focusing on one-shot magnitude pruning in static settings and compute-adaptive early exit in dynamic scenarios. The author establishes a concentration theorem for one-shot magnitude pruning in a single-neuron model, demonstrating that the cosine similarity between full and pruned weight vectors converges to a constant dependent on the pruning rate. Additionally, the paper introduces the conditional perceptron model for early exit, revealing that the excess generalization error decreases as a power of the compute gap, with the exponent increasing as the alignment between partial and full computations improves. The analysis extends to deep networks, where the accumulation of pruning-induced distortions is characterized, leading to a compute-accuracy tradeoff for frozen-backbone early exits modeled as a neural network Gaussian process. Numerical simulations validate the theoretical predictions, providing insights into the mechanisms governing compute reduction in neural networks from single neurons to deep architectures.
Methodology
The paper employs theoretical analysis starting from a single-neuron model to derive results on one-shot magnitude pruning and adaptive early exit. It uses concentration theorems and Gaussian process modeling to extend findings to deep networks, analyzing how pruning and early exit strategies affect compute and accuracy.
Results
The main results include a closed-form characterization of pruning distortion in single neurons, a power-law decay of excess generalization error in the conditional perceptron model, and a detailed analysis of how pruning distortions accumulate in deep networks, leading to a quantifiable compute-accuracy tradeoff.
Implications
The findings suggest that effective compute reduction strategies can be developed for neural networks, potentially leading to more efficient architectures that maintain performance while reducing computational costs. This has implications for deploying AI models in resource-constrained environments.
Sampling via Decision-Flow: Training-Free Extraction of Improved Latent Reasoning Paths in Large Language Models
NLP
Large Language Models
Reinforcement Learning
- DF-Sample is a training-free framework that enhances reasoning accuracy in LLMs.
- The method constructs a hierarchical reasoning tree and evaluates global trajectories instead of making local decisions.
- DF-Sample consistently outperforms traditional sampling strategies and RL-trained models across various benchmarks.
- The findings suggest that high-quality reasoning paths are latent in base models and can be accessed without additional training.
Read more
Sampling via Decision-Flow: Training-Free Extraction of Improved Latent Reasoning Paths in Large Language Models
Summary
This paper addresses the challenge of enhancing reasoning capabilities in large language models (LLMs) without the need for reinforcement learning (RL) fine-tuning. The authors propose a novel framework called Decision-Flow Sampling (DF-Sample), which operates at inference time and extracts high-quality reasoning paths that are latent within the base models. The core idea is based on the distribution-sharpening hypothesis, suggesting that RL reallocates probability mass towards high-reward paths that already exist in the model's distribution. DF-Sample constructs a hierarchical reasoning tree, evaluates terminal nodes for quality, and back-propagates utility scores to inform decision-making at intermediate branches. This approach allows for the recovery of reasoning chains that traditional sampling methods may overlook. The empirical results demonstrate that DF-Sample outperforms several baseline methods, including RL-trained models, across multiple benchmarks, indicating that substantial latent reasoning potential exists in pretrained models.
Methodology
The authors introduce DF-Sample, which constructs a hierarchical reasoning tree during inference. It evaluates the quality of terminal nodes and back-propagates these utility scores to inform the selection of reasoning paths. This global trajectory evaluation contrasts with traditional methods that rely on local sampling decisions.
Results
DF-Sample achieved 45.6% accuracy on the GPQA benchmark, surpassing power sampling (38.9%) and GRPO (39.9%). On the MATH500 dataset, it reached 81.8% accuracy, outperforming GRPO (78.5%) and power sampling (74.8%). The results across three model families and four datasets consistently demonstrate the effectiveness of DF-Sample compared to baseline methods.
Implications
The findings suggest that LLMs possess untapped reasoning capabilities that can be accessed without costly RL fine-tuning. This could lead to more efficient methods for enhancing model performance in various applications, including mathematics, coding, and scientific reasoning.
On-Device Language Models for Privacy-Preserving Stress Prediction: A Multimodal Evaluation on Mobile Health
NLP
Large Language Models
Multimodal
- ODLMs can effectively predict stress using multimodal data without cloud dependency.
- Objective sensor data marginally outperforms subjective self-reports in stress prediction.
- Lightweight models (sub-2B parameters) provide low latency and efficient resource usage.
- Zero-shot prompting strategies enhance the flexibility of stress prediction models.
Read more
On-Device Language Models for Privacy-Preserving Stress Prediction: A Multimodal Evaluation on Mobile Health
Summary
This paper investigates the use of on-device language models (ODLMs) for predicting stress levels in a privacy-preserving manner, leveraging multimodal data from mobile health applications. The authors highlight the importance of stress detection for mental health and the limitations of traditional supervised models that require retraining for new tasks. They propose using zero-shot prompting with ODLMs to analyze both objective sensor data and subjective self-reports, aiming to evaluate predictive accuracy, latency, and resource usage. The study finds that objective sensor features slightly outperform subjective reports, and lightweight models under 2 billion parameters achieve low latency and predictable resource usage. The research contributes to understanding how ODLMs can facilitate real-time stress monitoring while addressing privacy concerns associated with cloud-based models.
Methodology
The authors employed a systematic evaluation of selected ODLMs integrated into a mobile health platform. They utilized zero-shot prompting strategies to present time-series data, comparing different modalities and temporal serialization methods. The evaluation focused on predictive accuracy, latency, and throughput across various model sizes.
Results
The results indicate that ODLMs can achieve competitive predictive accuracy for stress prediction, with objective sensor data yielding better performance than subjective reports. Lightweight models demonstrated low latency and efficient resource usage, making them suitable for mobile health applications.
Implications
The findings suggest that ODLMs can be effectively utilized for real-time, privacy-sensitive stress monitoring in mobile health systems, paving the way for more accessible mental health interventions without compromising user privacy.
Reinforcement Learning over Patient Trajectories for Clinical Reasoning in EHR Foundation Models
Reinforcement Learning
Generative Models
Time Series
- Proposes RL fine-tuning for EHR foundation models to enhance clinical reasoning.
- Introduces time-aware, rollout-sensitive rewards for optimizing patient trajectory generation.
- Demonstrates that smaller models can outperform larger pre-trained models in data-limited settings.
- Shows positive transfer across multiple clinical reasoning tasks through multi-task RL.
Read more
Reinforcement Learning over Patient Trajectories for Clinical Reasoning in EHR Foundation Models
Summary
This paper addresses the limitations of Electronic Health Record (EHR) foundation models in clinical reasoning by proposing a reinforcement learning (RL) fine-tuning framework. Traditional EHR models, while effective in clinical prediction tasks, often rely on next-token prediction, which can be insufficient for long-horizon reasoning due to incomplete patient data. The authors introduce a novel approach that treats EHR models as generative policies over patient trajectories and formulates clinical prediction problems as event-conditioned, time-windowed reasoning tasks. They design time-aware, rollout-sensitive rewards to optimize the model's performance while considering finite rollout lengths and temporally inconclusive outcomes. The results demonstrate that RL fine-tuning significantly enhances the clinical reasoning capabilities of EHR models, allowing smaller models to outperform larger ones in data-limited scenarios and facilitating positive transfer across multiple clinical tasks. Additionally, the RL fine-tuned models generate patient trajectories that align more closely with ground truth, improving downstream predictive utility.
Methodology
The authors developed a reinforcement learning fine-tuning framework that treats pre-trained EHR foundation models as generative policies. They formulated clinical prediction tasks as event-conditioned, time-windowed reasoning problems and designed a reward system that accounts for the temporal dynamics of patient trajectories. The framework filters training data based on outcome relevance and rollout length, ensuring that the RL optimization process remains clinically meaningful.
Results
The RL fine-tuning approach consistently outperformed pre-trained models and strong baselines in both single-task and multi-task settings. Smaller models achieved superior performance compared to larger models in scenarios with limited data. Additionally, the generated patient trajectories exhibited stronger alignment with ground truth data and provided enhanced predictive utility when used in downstream classifiers.
Implications
This work has significant implications for improving clinical decision-making processes using EHR data. By enhancing the reasoning capabilities of EHR foundation models, the proposed framework can lead to better patient outcomes through more accurate predictions of clinical events. It also opens avenues for further research in optimizing clinical reasoning tasks using reinforcement learning techniques.
Toward Reliable Railway-Bogie Response Prediction Using Multifidelity TDNN and Physics-Informed Residual Learning
Time Series
- Introduces a multifidelity approach for railway-bogie response prediction.
- Combines low-fidelity simulation data with high-fidelity experimental measurements.
- Utilizes a time-delay neural network (TDNN) and a residual-correction network.
- Achieves significant accuracy in response predictions with a mean coefficient of determination of 0.8197.
Read more
Toward Reliable Railway-Bogie Response Prediction Using Multifidelity TDNN and Physics-Informed Residual Learning
Summary
This paper addresses the challenge of accurately predicting railway-bogie responses under various operating conditions, which are often impractical to test exhaustively. The authors propose a multifidelity approach that integrates low-fidelity multibody simulation data with high-fidelity roller-rig measurements. The methodology employs a time-delay neural network (TDNN) to capture the condition-dependent simulation trends, while a residual-correction network is used to model discrepancies between the simulated and physical responses. The approach is anchored in physics-informed learning, which constrains the model based on dynamic balance equations that account for differences in inertia, damping, stiffness, and external forces. The training objective combines several loss functions to ensure accurate response predictions. The results demonstrate a mean coefficient of determination of 0.8197 and a mean normalized root-mean-square error of 4.6055%, indicating a promising level of accuracy for the railway-bogie response predictions, particularly at high speeds.
Methodology
The methodology involves treating multibody simulation histories as low-fidelity data and roller-rig measurements as high-fidelity evidence. A TDNN is employed to model the baseline simulation trends, while a residual-correction network captures the discrepancies. The training incorporates physics-informed constraints based on dynamic balance equations, along with various loss functions to ensure smoothness and consistency in the predictions.
Results
The proposed method yields a mean coefficient of determination of 0.8197, a mean normalized root-mean-square error (NRMSE) of 4.6055%, and a mean normalized mean absolute error (NMAE) of 1.9297%. These results indicate a high level of accuracy in predicting railway-bogie responses, particularly validated at a speed of 385 km/h.
Implications
The findings suggest that the multifidelity approach can enhance the reliability of railway-bogie response predictions, which is crucial for the design and safety of railway systems. This methodology could be applied to other engineering domains where simulation and experimental data need to be integrated for accurate predictions.
TokenMapper: A Step Toward Interoperable Speech Token Translation
Audio & Speech
- TokenMapper enables direct token-to-token translation, eliminating the need for waveform decoding and re-encoding.
- The framework handles structural mismatches between different token spaces, including single and multi-codebook representations.
- Empirical results show that TokenMapper preserves intelligibility across different tokenizers with minimal WER.
- Significant reductions in latency (up to 94.5%) are achieved compared to conventional methods.
Read more
TokenMapper: A Step Toward Interoperable Speech Token Translation
Summary
The paper introduces TokenMapper, a novel framework designed to facilitate direct token-to-token translation between heterogeneous speech tokenizers, addressing the interoperability challenges faced in neural audio codecs. Traditional methods require decoding speech tokens to waveform audio before re-encoding them with a different tokenizer, leading to increased latency and potential information loss. TokenMapper operates entirely in the discrete domain, allowing for direct translation between token spaces that may differ in vocabulary and codebook structure. It supports both single and multi-codebook representations and maintains a shared effective token rate. The authors conducted experiments using three distinct speech tokenizers: GLM-4-Voice, MiMi, and DualCodec, demonstrating that TokenMapper can effectively translate tokens while preserving intelligibility. The results indicate that the translation word error rate (WER) is within 2.5-6.8% of native reconstructions, and the end-to-end latency is reduced by 4.8-94.5%, achieving up to 972 ms per utterance. This work represents a significant step towards enabling seamless communication between different speech models without the need for intermediate waveform reconstruction.
Methodology
TokenMapper employs a direction-aware mapping strategy to facilitate direct translation between different token spaces. The framework is designed to work with same-rate tokenizers, allowing for effective communication between models with varying vocabulary sizes and codebook structures. The authors evaluated the framework's performance through experiments on three distinct speech tokenizers, measuring translation accuracy and latency.
Results
The experiments demonstrated that TokenMapper achieves a translation WER of 5.85-9.98%, which is within 2.5-6.8% of the WER for native reconstructions. Additionally, the framework significantly reduces end-to-end latency by 4.8-94.5%, with reductions reaching up to 972 ms per utterance, showcasing its efficiency compared to traditional waveform bridging methods.
Implications
TokenMapper's ability to facilitate direct communication between different speech models without waveform reconstruction has significant implications for real-time applications such as conversational agents and speech-to-speech translation systems. This advancement could lead to more efficient and responsive speech processing technologies.
Observation-Anchored Selective Assimilation for Longitudinal Tumor-State Proxy Forecasting in Post-Treatment Glioma
Time Series
Computer Vision
- Introduction of Observation-Anchored Selective Assimilation (OASA) for tumor-state proxy forecasting.
- Utilization of longitudinal MRI data to enhance patient-specific tumor-state estimates.
- Comparison of OASA with multiple forecasting methods, showing competitive performance.
- Demonstration of improved calibration metrics alongside high Dice scores.
Read more
Observation-Anchored Selective Assimilation for Longitudinal Tumor-State Proxy Forecasting in Post-Treatment Glioma
Summary
This paper addresses the challenge of forecasting tumor-state proxies in post-treatment glioma patients using longitudinal MRI data. The authors propose a novel framework called Observation-Anchored Selective Assimilation (OASA), which integrates new follow-up observations to enhance patient-specific state estimates. The study analyzes data from 203 patients across 594 follow-up time points, applying a no-new-treatment criterion to retain relevant triplet data for training, validation, and testing. Each time point is represented by a voxel-wise tumor-state proxy map derived from MRI lesion labels. The OASA framework anchors the observed intermediate proxy and selectively applies model-predicted updates through a tiered case-level gating rule and voxel-wise soft gates. The performance of OASA is compared against various forecasting methods, demonstrating its effectiveness in maintaining high Dice scores while also improving calibration metrics. The results indicate that OASA can provide a more accurate and biologically relevant forecast of tumor-state proxies, highlighting the importance of incorporating intermediate observations in longitudinal forecasting tasks.
Methodology
The methodology involves a shared single-step forecaster that predicts updates from multimodal source-state tensors. The OASA framework anchors the observed intermediate proxy and applies selective updates based on a validation-selected tiered case-level rule and voxel-wise soft gates. The model uses a SegMamba-based encoder-decoder architecture to generate tumor-state proxy maps from MRI data.
Results
OASA maintained a Dice score comparable to persistence at τ = 0.2 (0.6071 ± 0.0025 vs. 0.6070) and achieved a higher Dice score at τ = 0.5 (0.4269 ± 0.0079 vs. 0.3981). Calibration improved Dice at τ = 0.2 to 0.6178 ± 0.0025, increased false-positive support, and reduced false-negative support, indicating enhanced near-threshold support calibration.
Implications
The findings suggest that incorporating intermediate observations in forecasting models can significantly improve the accuracy of tumor-state predictions in glioma patients. This approach may be applicable in other areas of medical imaging and longitudinal data analysis, potentially leading to better patient management and treatment planning.
Quantile-based Loss Filtering for Outlier-Robust Stochastic Gradient Descent
Optimization
Theory
- Introduction of Quantile-k-Loss SGD (QkL-SGD) for robust optimization against outliers.
- Linear convergence guarantees under convexity assumptions with sample size scaling based on corruptions.
- Probabilistic analysis for small sample sizes, linking convergence to outlier selection probability.
- Experimental results demonstrate superior performance of QkL-SGD compared to standard SGD and min-k-loss SGD.
Read more
Quantile-based Loss Filtering for Outlier-Robust Stochastic Gradient Descent
Summary
This paper introduces Quantile-k-Loss SGD (QkL-SGD), a novel framework for loss-based filtering in stochastic gradient descent (SGD) aimed at enhancing robustness against outliers in finite-sum optimization problems. The authors address the challenge posed by corrupted component functions whose gradients can significantly distort the optimization trajectory. QkL-SGD operates by sampling k component losses at each iteration and selecting an update index from the lower empirical q-quantile of these losses. The paper establishes linear convergence under standard convexity assumptions, with sample size requirements linked to the number of corruptions and a strong-convexity threshold. Additionally, a probabilistic analysis is provided for scenarios where large sampling is impractical, demonstrating that the convergence behavior is influenced by the likelihood of selecting outliers and the curvature of the selected good step. Experimental results on polynomial regression, logistic regression, and hinge loss indicate that QkL-SGD, particularly with intermediate quantiles, outperforms both standard SGD and min-k-loss SGD, highlighting its robustness and efficiency in producing informative updates.
Methodology
The methodology involves a two-parameter family of stochastic sampling rules where at each iteration, k component losses are sampled, and an update index is chosen uniformly from those losses that fall below the empirical q-quantile. The framework is analyzed for both deterministic and probabilistic convergence guarantees.
Results
The authors prove linear convergence of QkL-SGD under standard assumptions, with sample size requirements that depend on the number of corruptions rather than the total number of components. The experimental results show that QkL-SGD outperforms standard SGD and min-k-loss SGD, particularly in retaining robustness and providing informative updates.
Implications
The proposed QkL-SGD framework has potential applications in various fields where robust optimization is crucial, such as machine learning tasks involving noisy or corrupted data. It can enhance the reliability of SGD in real-world scenarios where data integrity cannot be guaranteed.
SCOPE-OPSD: Fisher-Conditioned Privileged Subspaces for On-Policy Self-Distillation
NLP
Large Language Models
Theory
- Introduces a final-layer privileged-residual objective that enhances OPSD without additional rollouts.
- Utilizes a Fisher-conditioned rank-64 subspace to isolate the effects of privileged information.
- Demonstrates consistent performance improvements across multiple model scales and checkpoints.
- Provides empirical evidence supporting the effectiveness of structured versus random orientations in training.
Read more
SCOPE-OPSD: Fisher-Conditioned Privileged Subspaces for On-Policy Self-Distillation
Summary
This paper introduces SCOPE-OPSD, a novel approach to On-Policy Self-Distillation (OPSD) that enhances the learning process by leveraging a Fisher-conditioned privileged subspace. The authors investigate the potential of utilizing the aligned final-layer discrepancies between a student model and a frozen self-teacher, aiming to improve the efficiency of knowledge transfer during training. The proposed method projects the privileged teacher-student residual onto a fixed low-rank factor derived from residual covariance and Fisher sensitivity, without adding extra inference-time components. The methodology is validated through extensive experiments on the Qwen3 model across various scales, demonstrating that the structured approach consistently outperforms traditional OPSD and matched random controls. The findings suggest that the structured privileged subspace effectively captures valuable training information, leading to improved model performance in on-policy distillation tasks.
Methodology
The methodology involves projecting the final-layer residuals of the student and privileged teacher through a fixed low-rank factor, which is estimated from covariance and Fisher sensitivity metrics. The structured approach is compared against a matched random control to isolate the effects of the learned orientation while maintaining the same rank and spectrum characteristics.
Results
The results indicate that the structured approach consistently outperforms pure OPSD and matched random controls across various model checkpoints, with significant improvements noted at step 75 on the Qwen3-1.7B model. Specifically, structured methods exceeded matched random by an average of 1.39 percentage points and showed a 4.40 times greater capture of the held-out privileged gap score.
Implications
The findings suggest that incorporating a Fisher-conditioned privileged subspace can enhance the efficiency of on-policy self-distillation, potentially leading to better performance in tasks requiring long-form reasoning and complex decision-making. This approach may have broader applications in training large language models and improving their reasoning capabilities.
Rank-Efficient LoRA via Joint Tangent-Space Optimization under Isotropic Curvature
NLP
Large Language Models
Optimization
- Effective rank utilization is crucial for optimizing LoRA adaptations.
- Different optimizers can significantly impact the effective rank achieved during training.
- Iso-LoRA optimizes LoRA updates by focusing on tangent-space updates, improving rank utilization.
- The proposed method shows substantial performance gains across various language models.
Read more
Rank-Efficient LoRA via Joint Tangent-Space Optimization under Isotropic Curvature
Summary
This paper addresses the optimization of Low-Rank Adaptation (LoRA) for large pretrained models, emphasizing the importance of effective rank utilization during the adaptation process. While LoRA traditionally uses a nominal rank to control the parameter budget and representational capacity, the authors argue that the optimizer plays a crucial role in determining how effectively this capacity is utilized. Through a case study on GPT-2 adaptation, they observe that different optimizers (AdamW vs. Muon) exhibit varying effects on rank utilization, with AdamW leading to concentrated singular spectra and low effective rank. To overcome this limitation, the authors propose Iso-LoRA, a geometry-aware optimizer that couples LoRA factor updates through spectral descent in the induced tangent space. Iso-LoRA encourages a more even distribution of energy across singular directions, enhancing effective rank and downstream performance. The paper provides theoretical guarantees for Iso-LoRA's effectiveness and validates its performance across various language models, demonstrating significant improvements, particularly at moderate-to-large LoRA ranks. This work highlights the critical interaction between optimizer design and rank utilization in parameter-efficient adaptation.
Methodology
The authors introduce Iso-LoRA, an optimizer that modifies the update process of LoRA factors by targeting the induced tangent update in weight space. It employs a spectral descent approach to ensure that updates distribute energy evenly across singular directions, rather than concentrating it in a few modes. The methodology includes a theoretical analysis of the optimizer's effectiveness and empirical validation through experiments on language model adaptations.
Results
Iso-LoRA was shown to improve effective rank and downstream performance across language models ranging from 0.1B to 7B parameters. The strongest performance gains were observed at moderate-to-large LoRA ranks, indicating that the proposed optimizer effectively utilizes the rank budget more efficiently than standard optimizers.
Implications
The findings suggest that optimizing the design of training algorithms can lead to better performance in parameter-efficient adaptations of large language models. This has potential applications in various NLP tasks where computational resources are limited, enabling more effective use of pretrained models.
SAGE-Loop: Reliable Closed-Loop LLM-Driven AutoML with Trial-and-Correction and Adaptive Ensembling
Large Language Models
Optimization
Interpretability
- SAGE-Loop introduces a closed-loop trial-and-correction mechanism for AutoML, enhancing reliability.
- The framework utilizes LLMs for adaptive model generation and ensemble strategies.
- Validation on 20 datasets shows consistent performance improvements across various tasks.
- SAGE-Loop can automatically repair execution failures, ensuring reliable run completion.
Read more
SAGE-Loop: Reliable Closed-Loop LLM-Driven AutoML with Trial-and-Correction and Adaptive Ensembling
Summary
The paper introduces SAGE-Loop, a novel framework for Automated Machine Learning (AutoML) that integrates large language models (LLMs) to enhance pipeline reliability and adaptive optimization. Traditional AutoML systems often lack mechanisms for real-time feedback and correction during execution, leading to suboptimal outcomes when errors occur. SAGE-Loop addresses this by implementing a closed-loop process that allows for trial-and-correction, enabling the system to adaptively refine its model generation and ensemble strategies based on execution feedback. The framework consists of three main components: feedbackable feature generation, trial-and-correction model generation, and adaptive ensemble strategies for both supervised and unsupervised tasks. The authors validate SAGE-Loop across 20 public datasets, demonstrating significant improvements in performance and stability in classification, regression, and clustering tasks. The results highlight SAGE-Loop's ability to recover from execution failures and maintain robust pipeline behavior, showcasing its potential to transform AutoML practices.
Methodology
SAGE-Loop employs a multi-round prompting strategy for model synthesis, treating LLMs as structure search agents. It includes feedbackable feature generation based on task patterns, trial-and-correction model generation that combines error-driven repairs with performance revisions, and an adaptive ensemble mechanism that selects appropriate strategies for both supervised and unsupervised tasks.
Results
The framework was validated on 20 public datasets, showing consistent gains in performance and stability across classification, regression, and clustering tasks. The results also indicated that SAGE-Loop effectively recovers from execution failures and maintains robust pipeline behavior, with ablation studies confirming the benefits of its closed-loop and diversity-utilization mechanisms.
Implications
SAGE-Loop's approach could significantly enhance the reliability and efficiency of AutoML systems, making them more accessible for non-experts and applicable in various data-driven fields. Its ability to adaptively correct and optimize models in real-time could lead to more effective machine learning solutions in practice.
Almost Sure Convergence Analysis of Stochastic Gradient Methods with Clipping and Additive Noise
Optimization
Theory
- SGD with clipping and additive noise converges almost surely under specific conditions.
- Clipping introduces bias, but the conditional mean of the clipped gradient remains a descent direction if the clipping threshold is sufficiently large.
- The analysis extends to momentum variants, providing similar convergence guarantees.
- The results suggest stability of SGD-CN in both convex and nonconvex optimization problems.
Read more
Almost Sure Convergence Analysis of Stochastic Gradient Methods with Clipping and Additive Noise
Summary
This paper investigates the convergence properties of Stochastic Gradient Descent (SGD) methods that incorporate gradient clipping and additive Gaussian noise, referred to as SGD with Clipping and Additive Noise (SGD-CN). The authors establish that SGD-CN converges almost surely under certain conditions, specifically when the stochastic gradient noise is uniformly bounded and the objective function is smooth. They highlight that while gradient clipping introduces bias and additive noise increases variance, the algorithm can still maintain stability in both convex and nonconvex optimization scenarios. The analysis extends to momentum-based variants of SGD, such as the stochastic heavy ball and Nesterov's accelerated gradient methods, demonstrating that similar convergence guarantees can be achieved through careful energy constructions. The findings provide a stronger theoretical foundation for understanding the behavior of clipped stochastic gradient methods, suggesting that they can be effectively utilized in applications requiring robustness and privacy.
Methodology
The authors utilize a theoretical framework based on supermartingale convergence and weighted summability to analyze the convergence of SGD-CN. They establish conditions under which the stochastic gradient noise is uniformly bounded and the objective function is smooth, leading to almost sure convergence results. The methodology includes deriving properties of clipped gradients and controlling oscillations in the gradient norm to strengthen convergence guarantees.
Results
The paper proves that SGD-CN converges almost surely under the assumptions of smoothness and uniformly bounded stochastic gradient noise. It establishes that the conditional mean of the clipped gradient can still serve as a descent direction, provided the clipping threshold is appropriately chosen. The analysis also shows that the convergence results extend to momentum-based methods, ensuring stability in optimization trajectories.
Implications
The findings have significant implications for the design of robust and privacy-preserving optimization algorithms in machine learning. The theoretical guarantees provided by the authors can enhance the reliability of SGD methods in practical applications, particularly in scenarios where gradient clipping and noise are necessary for stability and privacy.
A Unified and Constrained View of Regularization-Based Robust Reinforcement Learning
Reinforcement Learning
Optimization
Theory
- Unification of regularization-based methods for robust DRL through new performance gap bounds.
- Introduction of a constrained optimization framework for robust training that adapts regularization weight.
- Empirical validation of theoretical analysis through extensive adversarial evaluations.
- Demonstration of improved robustness when combining different regularization techniques.
Read more
A Unified and Constrained View of Regularization-Based Robust Reinforcement Learning
Summary
This paper addresses the challenge of training robust Deep Reinforcement Learning (DRL) policies against adversarial input perturbations by unifying various regularization-based methods. The authors derive new upper bounds on the performance gap between nominal and worst-case policies, demonstrating that these bounds can be expressed as a combination of existing regularization objectives and a KL-divergence penalty. This formulation elucidates the empirical effectiveness of adding a KL penalty for robustness. The authors propose a novel approach to robust training as a constrained optimization problem, allowing for the dynamic adjustment of the regularization weight during training. Extensive evaluations on continuous control tasks validate the theoretical findings, showing that the proposed method improves robustness compared to existing approaches.
Methodology
The authors derive upper bounds on the performance gap between nominal and worst-case policies, reformulating robust training as a constrained optimization problem. They utilize existing regularization methods and introduce a dynamic Lagrange multiplier that adjusts the regularization weight based on the current robustness of the policy.
Results
The proposed method shows significant improvements in robustness across various continuous control tasks when compared to traditional fixed regularization approaches. The theoretical analysis is supported by empirical results demonstrating the effectiveness of the unified framework.
Implications
This work has implications for the development of more robust DRL systems, particularly in environments where adversarial perturbations are a concern. The dynamic adjustment of regularization weights could lead to more efficient training processes and better performance in real-world applications.
ParaRecover: A Process-Level Benchmark for Error Localization and Recovery in Parallel Tool-Use Agents
NLP
Large Language Models
Reinforcement Learning
- Introduction of ParaRecover, a benchmark for evaluating error localization and recovery in parallel tool-use agents.
- Development of a fine-grained error taxonomy covering 14 error types relevant to multi-turn parallel execution.
- Proposal of the SDE rubric for detailed assessment of agent behavior during error recovery.
- Experiments reveal significant weaknesses in current state-of-the-art models regarding error propagation and recovery.
Read more
ParaRecover: A Process-Level Benchmark for Error Localization and Recovery in Parallel Tool-Use Agents
Summary
The paper introduces ParaRecover, a novel benchmark designed to evaluate error localization and recovery capabilities in multi-turn parallel tool-use agents. Traditional benchmarks primarily focus on final task success or tool-call correctness, which do not adequately assess an agent's ability to diagnose and recover from intermediate execution failures. This is particularly important in multi-turn scenarios where errors can propagate and cause cascading failures. ParaRecover features a fine-grained taxonomy of 14 error types related to planning dependencies, tool selection, and argument matching, encompassing 10,626 instances across two difficulty levels. The authors propose the SDE rubric, which evaluates agents based on structural integrity, diagnostic reasoning, and evolutionary strategy during execution. Experiments with over ten mainstream large language models (LLMs) reveal that even state-of-the-art models struggle with multi-turn error propagation and precise replanning. The SDE rubric is shown to provide effective supervision signals for enhancing agents' reflective recovery capabilities, highlighting a significant gap between task completion rates and process reliability in current agents.
Methodology
The authors constructed a benchmark consisting of 10,626 instances categorized into two difficulty levels. They developed a fine-grained error taxonomy based on real agent execution trajectories and proposed the SDE evaluation rubric to assess agents' performance across three dimensions: Structural Integrity, Diagnostic Reasoning, and Evolutionary Strategy. Experiments were conducted using various mainstream LLMs to evaluate their performance on the benchmark.
Results
The results indicate that current state-of-the-art models, despite achieving high task completion rates, exhibit significant deficiencies in dependency reasoning, implicit failure localization, and efficient recovery from multi-turn error propagation. The SDE rubric was effective in providing supervision signals that improved agents' reflective recovery capabilities.
Implications
The findings suggest that existing benchmarks for agent performance are insufficient for understanding the reliability of agents in complex, real-world tasks. ParaRecover can serve as a critical tool for researchers to develop more robust agents capable of handling errors in multi-turn parallel scenarios, which is essential for applications in fields like scientific research, code generation, and data analysis.
Physical-State-Guided Diffusion Sampling for Full-Waveform Inversion
Generative Models
Optimization
Theory
- PSG effectively combines physical velocity with a learned geological prior to enhance FWI.
- The method preserves conventional initialization and optimization history while improving reconstruction accuracy.
- PSG outperforms classical and diffusion-based baselines in various acquisition scenarios.
- Ensemble variability in results is concentrated near geological interfaces, indicating a relationship with inversion errors.
Read more
Physical-State-Guided Diffusion Sampling for Full-Waveform Inversion
Summary
This paper introduces Physical-State-Guided Diffusion Sampling (PSG) as a novel approach to Full Waveform Inversion (FWI), which aims to estimate subsurface velocity from seismic recordings. The authors address the challenges of ill-posedness and nonlinearity in FWI, which make accurate reconstruction heavily reliant on initialization and prior information. PSG integrates a persistent physical velocity with a diffusion prior using a Gaussian bridge, refining the physical state through waveform fitting that is regularized by the denoised velocity. This method effectively separates the gradients of the wave-equation and denoiser while preserving the conventional FWI initialization and optimization history. The performance of PSG is evaluated on four OpenFWI families, demonstrating that its terminal denoised estimates surpass classical and diffusion-based methods, even under conditions of clean and missing-trace acquisitions, while maintaining strong structural recovery amidst measurement noise. The approach also shows that repeated stochastic runs can preserve dominant geological structures, with ensemble variability closely linked to local inversion errors. Furthermore, a frozen OpenFWI-trained prior facilitates the inversion of complex geological structures in larger models without the need for retraining.
Methodology
The methodology involves coupling a persistent physical velocity to a diffusion prior through a Gaussian bridge. The physical state is refined via waveform fitting, which is regularized by the denoised velocity, guiding the reverse diffusion process. This separation of gradients allows for effective optimization while leveraging learned geological priors.
Results
PSG's terminal denoised estimates significantly outperform both classical and diffusion-based methods across various scenarios, including clean and missing-trace acquisitions. The method maintains strong structural recovery under measurement noise and demonstrates the ability to recover complex geological structures in larger models without retraining.
Implications
The proposed PSG method has significant implications for seismic imaging and subsurface exploration, providing a robust framework for accurately estimating geological structures from seismic data. It enhances the reliability of FWI in practical applications, potentially benefiting resource exploration and environmental monitoring.
Benign Loss Landscapes Can Coexist with Worst-Case Hardness
Theory
Optimization
- TTNs can express arbitrary read-once Boolean formulas, including hard-to-learn targets.
- All minimum-norm local minima in TTNs are global minima, indicating benign loss landscapes.
- Learning difficulty in TTNs is attributed to high-order degenerate saddle points caused by rank-deficiency.
- The paper provides a case study on the parity function to illustrate the relationship between landscape geometry and computational hardness.
Read more
Benign Loss Landscapes Can Coexist with Worst-Case Hardness
Summary
This paper investigates the structure of loss landscapes in deep learning, specifically focusing on tree tensor networks (TTNs). The authors demonstrate that TTNs can express complex targets, including those that are hard to learn via gradient descent, such as the parity function. Despite this complexity, the authors prove that TTNs exhibit benign loss landscapes, where all local minima that are minimum-norm are also global minima. This finding challenges the conventional understanding of learning difficulty, suggesting that the challenges in training TTNs arise from high-order degenerate saddle points rather than bad local minima. The paper provides a detailed analysis of these saddle points, linking them to rank-deficiency in the model. The authors argue that TTNs can serve as a new surrogate model for studying the dynamics of deep learning training, offering insights into how the computational structure of targets influences optimization landscapes.
Methodology
The authors utilize tree tensor networks (TTNs) as a model class that generalizes deep linear networks. They analyze the loss landscapes of TTNs and prove theoretical results regarding the nature of local minima and saddle points. The study includes a case analysis of the parity function to demonstrate the implications of their findings.
Results
The authors prove that TTNs lack bad local minima under a minimum-norm condition, and that all local minima that are minimum-norm are globally optimal. They also show that non-optimal parameter points can lead to continuous non-increasing loss paths, indicating no 'bad valleys' exist in the loss landscape. Furthermore, they characterize high-order degenerate saddle points linked to rank-deficiency, which contribute to learning difficulties.
Implications
The findings suggest that TTNs could be a valuable tool for understanding the optimization dynamics in deep learning, particularly in relation to the structure of loss landscapes and the nature of learning difficulties. This could lead to improved training strategies and insights into the behavior of neural networks in practice.
Temporal Recurrence Favors Fewer Layers
Reinforcement Learning
NLP
Efficient ML
- Temporal recurrence allows for fewer layers in recurrent models while achieving comparable or better performance.
- The study formulates the problem as a compute-allocation challenge, comparing recurrent and non-recurrent models.
- Increasing parallel capacity benefits both model types, but gains from additional layers saturate earlier with recurrence.
- The findings suggest a broader hypothesis that temporal recurrence may favor shallower updates in other streaming contexts.
Read more
Temporal Recurrence Favors Fewer Layers
Summary
This paper investigates the impact of temporal recurrence on the depth of recurrent models in streaming tasks. The authors frame the question of how much within-step depth is necessary when temporal recurrence is available, suggesting that it can allow for fewer layers while maintaining or improving performance. They approach this as a compute-allocation problem, varying the within-step depth, expert width, and the number of parallel experts across different compute budgets. The study finds that recurrent models can achieve comparable or better performance with significantly fewer layers compared to non-recurrent models. This shift in compute allocation suggests that temporal recurrence allows for more efficient use of computational resources, emphasizing the importance of parallel computation over depth in certain contexts. The findings are validated through experiments on Sokoban and autoregressive FineWeb language modeling, demonstrating that recurrence reduces the performance cost associated with fewer layers and highlights the distinct roles of recurrence and parallel capacity.
Methodology
The authors conducted experiments comparing recurrent and non-recurrent models by varying the within-step depth, expert width, and the number of parallel experts under matched computational budgets. They analyzed the performance of these models on tasks such as Sokoban and FineWeb language modeling.
Results
The results indicate that recurrent models can effectively utilize fewer layers for similar or improved performance compared to non-recurrent models. The best observed recurrent allocations used significantly fewer layers, while the performance benefits from increasing expert count were consistent across both model types.
Implications
These findings suggest that in streaming tasks where temporal recurrence is applicable, models can be designed with fewer layers, potentially leading to reduced latency and more efficient computation. This could influence the design of future architectures in various domains, including language modeling and reinforcement learning.
DCRA: Diffusion-Conditioned Representation Alignment for Robust Time-Series Learning
Time Series
- Introduction of DCRA framework leveraging forward diffusion for structured corruption in time-series learning.
- Feature-level consistency mechanism enhances representation alignment between clean and corrupted signals.
- DCRA shows improved robustness and sensitivity in seizure detection tasks on the CHB-MIT EEG dataset.
- Framework is encoder-agnostic, compatible with various model architectures.
Read more
DCRA: Diffusion-Conditioned Representation Alignment for Robust Time-Series Learning
Summary
The paper presents a novel framework called Diffusion-Conditioned Representation Alignment (DCRA) aimed at improving the robustness of time-series representation learning, particularly in clinical applications like EEG and ECG analysis. The authors address the challenges posed by noise and distribution shifts in time-series data, which often lead to unstable feature representations. DCRA utilizes the forward diffusion process as a structured corruption scheduler, allowing for a controlled evolution of representations across varying noise levels. This approach contrasts with traditional augmentation methods that treat noise as independent perturbations. The framework introduces a feature-level consistency objective that aligns representations from clean and corrupted signals, ensuring that the latent features remain semantically coherent and class-discriminative. The encoder-agnostic nature of DCRA allows it to be integrated with various architectures, including state space models and Transformers. Experimental results on the CHB-MIT EEG dataset demonstrate that DCRA significantly enhances performance under multiple noise conditions, achieving higher sensitivity at lower false-positive rates. The findings suggest that structured corruption combined with representation alignment is effective for robust time-series learning.
Methodology
The DCRA framework employs the forward diffusion process to create a structured corruption scheduler for time-series data. It introduces a feature-level consistency objective that aligns representations from clean and corrupted signals, promoting robust and semantically coherent latent features. The framework is designed to be compatible with various encoder architectures, including state space models and Transformers.
Results
The experimental evaluation on the CHB-MIT EEG dataset reveals that DCRA consistently outperforms baseline models under various noise conditions, achieving higher sensitivity while maintaining low false-positive rates. The analysis indicates that DCRA produces more balanced and structured representations compared to traditional and diffusion-only models.
Implications
The DCRA framework has significant implications for clinical applications where robust time-series analysis is critical, such as in seizure detection and other medical signal processing tasks. Its ability to maintain performance under noise and distribution shifts could enhance the reliability of automated diagnostic systems.
Dual-guided Hierarchical Edge Localization for Large-scale Optimal Transport Across Dimensions
Optimization
Efficient ML
Theory
- HELLO offers a scalable solution for large-scale discrete optimal transport problems.
- The dual-guided approach enhances both initialization and refinement processes.
- Achieves significant runtime improvements and lower transport objectives at large scales.
- Proven finite termination at a global optimum under specific conditions.
Read more
Dual-guided Hierarchical Edge Localization for Large-scale Optimal Transport Across Dimensions
Summary
This paper introduces HELLO, a novel hierarchical solver for large-scale discrete optimal transport (OT) that addresses the computational challenges posed by traditional methods. The authors propose a dual-guided edge localization approach that utilizes dual potentials for both coarse-to-fine initialization and within-level refinement of transport plans. By propagating coarse dual potentials through a recursive subsampling hierarchy, HELLO efficiently assigns candidate edges for transport. The refinement process iteratively incorporates the largest dual violators until the KKT residual meets a specified tolerance, while ensuring linear memory complexity through budgeted pruning. The method achieves significant runtime improvements, demonstrating lower transport objectives at the million-point scale and effectively scaling to 1.28 million samples in high-dimensional spaces. The framework also extends beyond standard discrete OT to support various pairwise costs, making it a versatile tool for applications in semi-discrete OT, Gromov-Wasserstein, unbalanced OT, and OT-based Flow Matching.
Methodology
The methodology involves a dual-guided hierarchical edge localization strategy. Coarse dual potentials are propagated across a recursive subsampling hierarchy to initialize candidate edges. The refinement phase iteratively adds dual violators while maintaining a budgeted pruning strategy to ensure linear memory usage. The process continues until the relative KKT residual is below a defined threshold, ensuring convergence to an optimal solution.
Results
HELLO demonstrates order-of-magnitude runtime improvements over existing baselines, achieving lower transport objectives at the million-point scale. It successfully scales to 1.28 million samples per marginal in 8192 dimensions, utilizing 41.6 GiB of peak GPU memory while maintaining a full relative KKT residual below 10^-6.
Implications
The proposed framework has significant implications for various applications in machine learning, including domain adaptation, computational biology, and generative modeling. Its scalability and efficiency make it suitable for handling large datasets and high-dimensional spaces, potentially transforming how optimal transport problems are approached in practice.
Write on Paper and Get the Online Digital Trace: A New Era for Handwriting
Robotics
Efficient ML
Multimodal
- Introduction of Digipen, a digital pen that captures handwriting on regular paper.
- Combination of hardware and software innovations to reconstruct handwriting traces using inertial sensors.
- Utilization of deep learning algorithms for real-time handwriting trace reconstruction.
- Focus on enhancing learning experiences by allowing handwriting on paper while capturing digital traces.
Read more
Write on Paper and Get the Online Digital Trace: A New Era for Handwriting
Summary
This paper presents an innovative solution for digitizing handwriting on regular paper using a digital pen equipped with inertial sensors. The authors address the challenge of capturing the pen's trajectory without requiring a specialized substrate or external reference systems. The proposed system integrates a sensor-equipped digital pen, named Digipen, with advanced artificial intelligence algorithms to reconstruct handwriting traces in real-time. The Digipen can function as a regular pen while simultaneously tracking movements through accelerometers and gyroscopes. The authors developed a deep learning model to process the inertial data and reconstruct the handwriting trace, optimizing the AI algorithms for efficient execution on the pen's embedded system. This work aims to bridge the gap between traditional handwriting and digital capture, enhancing learning experiences and providing a low-cost solution for educational settings.
Methodology
The authors developed a digital pen equipped with inertial sensors to track pen movements. They implemented deep learning techniques to process the inertial data for handwriting reconstruction. The system was designed to operate in real-time, optimizing AI algorithms for embedded execution. A special version of the Digipen was created to collect training data by recording both pen traces and inertial data simultaneously.
Results
The proposed system successfully reconstructs handwriting traces from IMU signals with high accuracy. The integration of deep learning models allowed for effective processing of noisy inertial data, demonstrating the feasibility of digitizing handwriting on paper without the need for specialized substrates. The system showed promise for educational applications, particularly in improving handwriting proficiency among children.
Implications
This research has significant implications for educational technology, providing a practical solution for digitizing handwriting in classrooms. It allows for the benefits of traditional handwriting practices while enabling digital capture and analysis, potentially enhancing learning outcomes. The technology could also be adapted for various applications in note-taking, document annotation, and handwriting recognition.
Explainable Prediction from Mobile Sensing Data through LLM-guided Concept Integration
NLP
Large Language Models
Multimodal
- Introduction of the Concept-Integrated Transformer (CIT) framework for explainable health predictions.
- Utilization of LLM for annotation-free concept abnormality supervision.
- High performance on two longitudinal datasets, achieving F1 scores of 0.756 and 0.765.
- Demonstration of interpretable behavioral patterns through learned concept scores.
Read more
Explainable Prediction from Mobile Sensing Data through LLM-guided Concept Integration
Summary
This paper presents a novel framework called Concept-Integrated Transformer (CIT) designed for explainable health-related predictions using mobile sensing data. The authors address the challenges of limited outcome supervision and the need for interpretability in small-cohort health studies. CIT leverages a pretrained large language model (LLM) to generate concept abnormality targets without requiring manual annotations. This allows the model to learn from longitudinal multimodal sensing data while providing interpretable outputs that reflect meaningful behavioral and physiological patterns. The framework consists of two modules: one for LLM-guided concept supervision and another for concept-supervised prediction. The authors evaluate CIT on two datasets, achieving high F1 scores and demonstrating that the learned concept scores can reveal significant behavioral insights, such as differences in sleep patterns between groups with varying levels of negative affect. Overall, the study highlights the potential of LLM-guided concept integration for enhancing predictive accuracy and interpretability in health-related predictions from mobile sensing data.
Methodology
The CIT framework comprises two modules: one for LLM-guided concept abnormality supervision and another for concept-supervised prediction. The LLM evaluates baseline-aware statistical summaries to produce concept-level abnormality targets with confidence weights, which guide the Transformer model in predicting health-related outcomes and concept scores from mobile sensing data.
Results
CIT achieved the highest F1 score of 0.756 on the AFFECT dataset and tied for the highest score of 0.765 on the PHQ-9 dataset. The learned concept scores provided interpretable insights into behavioral and physiological differences, particularly highlighting the impact of sleep quantity and quality on negative affect.
Implications
The findings suggest that LLM-guided concept integration can significantly enhance the accuracy and interpretability of health-related predictions from mobile sensing data, making it a valuable approach for small-cohort studies where traditional outcome labels are sparse.
MAxBench: A Multinomial Concept Recovery Benchmark
NLP
Large Language Models
Interpretability
- Introduces MAxBench, a framework for evaluating multinomial concept representations.
- Demonstrates that affine subspaces outperform other geometries in terms of reliability and recall.
- Finds that better non-zero offsets contribute significantly to the advantages of affine subspaces.
- Shows that manifold steering can be competitive but does not consistently outperform prompting.
Read more
MAxBench: A Multinomial Concept Recovery Benchmark
Summary
The paper introduces MAxBench, a geometry-agnostic evaluation framework designed for multinomial concept representations in language models. Unlike binary concepts that can be represented by a single direction in activation space, multinomial concepts (e.g., ANIMALS, COUNTRIES) require more complex representations due to their inherent subcategories. The authors explore various localization methods across different geometries and models to assess their effectiveness in recovering these representations. The study reveals that affine subspaces provide more reliable steering and higher recall compared to rank-one or linear subspaces, with much of the advantage attributed to better non-zero offsets. Additionally, manifold steering shows competitive performance when applicable, but no method consistently surpasses prompting. The findings highlight the need for broader interpretability research that accommodates the complexity of multinomial concepts.
Methodology
The authors developed a sampling-based steering framework to evaluate concept representations, which is geometry-agnostic. They compared 10 localization methods across 6 concepts, 5 geometries, and 4 models, measuring the quality of representations based on exclusivity and completeness.
Results
The study found that affine subspaces are more effective for steering multinomial concepts, achieving higher recall rates. The advantage of affine methods is largely due to improved non-zero offsets rather than the choice of bases. Manifold steering is competitive when applicable, but prompting remains the most effective method overall.
Implications
The findings suggest that future interpretability research should focus on complex, multinomial concepts, potentially leading to improved steering methods in language models. This could enhance the control and understanding of model behaviors in various applications.
DynSHAP: Towards Explainable Dynamic Survival Analysis
Time Series
Interpretability
- DynSHAP extends SHAP to dynamic survival analysis by discretizing time into time-feature pairs.
- Temporal DynSHAP improves the recovery of temporally dependent features compared to traditional methods.
- The framework is validated on both synthetic and real-world clinical datasets, demonstrating practical applicability.
- DynSHAP provides insights into the timing of feature influence on survival predictions, enhancing interpretability.
Read more
DynSHAP: Towards Explainable Dynamic Survival Analysis
Summary
The paper introduces DynSHAP, a novel framework designed to enhance explainability in dynamic survival analysis (DSA) by adapting the SHAP (SHapley Additive exPlanations) methodology to handle longitudinal patient data effectively. Traditional deep learning models in DSA have shown strong predictive capabilities but are often criticized for their black box nature, which hinders clinical trust and adoption. Existing explainability methods fail to accommodate the complexities of longitudinal, irregular inputs and functional survival outputs, creating a significant gap in usability for clinical applications. DynSHAP addresses this by treating time-feature pairs as players in the Shapley game, allowing for a more nuanced understanding of feature importance over time. The framework includes Temporal DynSHAP, which captures linear dependencies among features across time and utilizes conditional sampling to enhance the accuracy of attributions. The authors validate DynSHAP on synthetic datasets with known ground-truth attributions, demonstrating its superior performance in recovering temporally dependent features compared to traditional marginal estimators. Furthermore, the framework is applied to two real-world clinical datasets, where it successfully produces attributions that align with established clinical knowledge, thereby facilitating a clearer understanding of the factors influencing survival predictions.
Methodology
DynSHAP modifies the SHAP framework to accommodate dynamic survival analysis by treating time-feature pairs as Shapley players. It introduces Temporal DynSHAP, which learns and accounts for linear dependencies among features over time, employing conditional sampling to improve the accuracy of feature attributions.
Results
The application of DynSHAP to synthetic data showed improved accuracy in recovering ground-truth attributions for temporally dependent features compared to marginal SHAP estimators. In real-world clinical applications, DynSHAP produced attributions that were consistent with established clinical findings, thereby validating its effectiveness and reliability.
Implications
DynSHAP has significant implications for the medical field, as it enhances the interpretability of dynamic survival analysis models, allowing clinicians to understand the underlying factors influencing patient outcomes. This increased transparency can foster greater trust in AI-driven predictions and facilitate more informed clinical decision-making.
Quality-Constrained Routing over a Fixed Pool of Quantized Mixture-of-Experts Instances
Optimization
Large Language Models
Efficient ML
- Introduces a fixed-pool routing problem for quantized MoE instances with a focus on quality constraints.
- Develops Fragility-Weighted Perplexity (FWP) as a predictive metric for request-specific risk.
- Demonstrates that FWP allocation significantly improves throughput compared to static and request-agnostic methods.
- Establishes a two-timescale framework separating provisioning decisions from routing actions.
Read more
Quality-Constrained Routing over a Fixed Pool of Quantized Mixture-of-Experts Instances
Summary
This paper addresses the challenge of routing requests to a fixed pool of quantized Mixture-of-Experts (MoE) instances, aiming to maximize throughput while adhering to quality-degradation budgets. The authors introduce a novel metric, Fragility-Weighted Perplexity (FWP), which quantifies the request-specific risk associated with instance degradation due to quantization. The methodology involves a two-timescale approach where provisioning decisions regarding instance configurations are made separately from the routing decisions. The routing problem is formalized as a linear programming (LP) problem that optimizes token serving under quality constraints. Empirical evaluations on 88 extended Qwen prompts demonstrate that the proposed FWP allocation method outperforms both static and request-agnostic mixing strategies, achieving a 1.284× offline model-based multiplier compared to 1.253× for request-agnostic mixing and 1.000× for static W4 instances. The findings highlight the importance of adaptive routing strategies in optimizing resource allocation in MoE systems.
Methodology
The authors employ a two-timescale approach to separate provisioning decisions from routing actions. They introduce Fragility-Weighted Perplexity (FWP) to assess request-specific risks based on expert fragility and affinity. A window-level linear programming model is formulated to maximize token serving while adhering to quality-degradation constraints, with the dual formulation ensuring KKT-consistency with optimal routing.
Results
The empirical results indicate that the FWP allocation method achieves a 1.284× offline model-based multiplier on the Qwen prompts, outperforming request-agnostic mixing (1.253×) and static W4 instances (1.000×). The FWP method contributes an additional 2.5% relative gain over request-agnostic mixing, demonstrating its effectiveness in optimizing resource allocation.
Implications
The findings suggest that adaptive routing strategies based on request-specific risk can significantly enhance the performance of quantized MoE systems. This approach can be applied in various applications requiring efficient resource management and quality control in machine learning models.
FINESSE: An Agent-Based Simulator and Benchmark Dataset for Multimodal Financial Event Sequences
Multimodal
Time Series
- FINESSE provides a novel agent-based model for generating synthetic, structured financial event sequences.
- The benchmark dataset, FINESSE-Bench, supports four representative financial tasks.
- Baseline results are reported using state-of-the-art methods, highlighting the effectiveness of the framework.
- The simulator addresses privacy concerns by generating data independent of sensitive information.
Read more
FINESSE: An Agent-Based Simulator and Benchmark Dataset for Multimodal Financial Event Sequences
Summary
The paper introduces FINESSE, a Financial Event Sequence Simulation Environment designed to address the limitations of existing financial datasets, which are often monomodal and narrow in focus. FINESSE employs an agent-based simulation framework to generate synthetic datasets that reflect the complex, interdependent nature of financial behaviors, such as transactions, payments, and account status changes. The authors present FINESSE-Bench, a benchmark dataset that supports four key tasks: balance forecasting, transaction fraud detection, missed payment prediction, and next event prediction. By utilizing methods from time series forecasting, event sequence modeling, temporal graphs, and temporal point processes, the paper reports baseline results for these tasks. The framework and dataset are made publicly available to facilitate further research in structured, multimodal event sequence modeling within the financial services sector.
Methodology
The authors developed an agent-based simulation framework (FINESSE-ABM) that generates synthetic datasets capturing interdependent event sequences in financial services. The simulation incorporates latent state dynamics to model the evolving behaviors of agents in response to various financial events. The generated datasets are then used to benchmark four tasks, employing various machine learning methodologies including time series forecasting and event sequence modeling.
Results
The paper provides initial baseline results for the four tasks supported by FINESSE-Bench, demonstrating the potential of the synthetic datasets for evaluating modeling approaches in financial services. The results indicate that the framework can effectively simulate complex financial interactions and provide valuable insights for research.
Implications
FINESSE has the potential to significantly enhance research in financial services by providing a rich, structured dataset that reflects real-world complexities. It can facilitate the development of more robust models for predicting financial behaviors and detecting fraudulent activities, ultimately contributing to improved decision-making in the industry.
Certifying Concept Unlearning in Text-to-Image Diffusion Models
Generative Models
Computer Vision
Theory
- Introduces a certification framework for concept unlearning in T2I diffusion models.
- Demonstrates that existing ASR metrics do not reliably bound residual leakage.
- Establishes upper bounds on leakage probabilities across a continuous concept neighborhood.
- Evaluates the framework on NSFW content, artistic styles, and celebrity identities.
Read more
Certifying Concept Unlearning in Text-to-Image Diffusion Models
Summary
This paper addresses the limitations of existing evaluations of concept unlearning in text-to-image (T2I) diffusion models, which primarily rely on attack success rates (ASR) from adversarial prompt searches. The authors argue that these metrics fail to provide a comprehensive understanding of residual concept leakage, potentially leading to overestimations of unlearning effectiveness and underestimations of safety risks. To overcome this, they introduce a novel certification framework that offers high-confidence guarantees on bounded error for residual concept leakage. This framework combines statistical certification with worst-case analysis across concept-relevant embedding directions, allowing for explicit upper bounds on leakage probabilities. The authors evaluate their framework across three major concept categories—NSFW content, artistic styles, and celebrity identities—using six state-of-the-art unlearning methods. Their findings reveal that certified leakage bounds exceed standard ASR by an average of 16.2%, highlighting significant residual risks overlooked by previous evaluation protocols. The results underscore the necessity of certification as a complementary tool for reliable auditing of concept unlearning in T2I models.
Methodology
The authors developed a certification pipeline that combines adversarial steering in a continuous concept neighborhood with calibrated text-space and pixel-space verification. This approach allows for the determination of whether a concept can still be elicited and whether it persists in generated outputs. The framework derives statistical certificates that provide upper bounds on the probability of regenerating an unlearned concept, rather than relying on a finite set of prompts.
Results
The evaluation of the certification framework across three concept categories and six unlearning methods showed that certified leakage bounds consistently exceeded standard ASR metrics by an average of 16.2%. This indicates that traditional attack-based evaluations significantly underestimate residual leakage, emphasizing the importance of the proposed certification method.
Implications
The findings suggest that the certification framework can enhance the safety and reliability of T2I diffusion models by providing a more comprehensive assessment of concept unlearning. This has implications for applications in content moderation, copyright protection, and privacy, where ensuring the suppression of sensitive concepts is critical.
GSF-χ: Global Stereochemical Fields for Chiral Graph Transformers
Graph Learning
- GSF-χ introduces a global stereochemical field that modulates pairwise interactions among all atoms.
- The Chiral-RoPE operator allows for reflection-invariant attention mechanisms that respect molecular chirality.
- The model achieves significant improvements in accuracy for chirality classification and ECD predictions.
- Extensive validation through property tests confirms the model's robustness and adherence to chirality principles.
Read more
GSF-χ: Global Stereochemical Fields for Chiral Graph Transformers
Summary
The paper introduces GSF-χ, a novel graph transformer designed to effectively handle molecular representations of chiral compounds. Traditional molecular encoders often struggle with enantiomers, which share identical atomic structures but exhibit different behaviors in chiral environments. GSF-χ addresses this by implementing a global stereochemical field that modulates interactions among all atoms based on stereogenic units, ensuring that the encoder respects atom relabelings and proper rotations while avoiding reflection biases. The core innovation is the Chiral-RoPE (Rotary Positional Encoding), which utilizes a pseudoscalar to direct relative rotations in the attention mechanism. This allows for a reflection-even phase field that maintains the architectural roles of all atoms. The authors validate their approach through extensive property tests and demonstrate that GSF-χ achieves state-of-the-art performance on chirality classification benchmarks, outperforming existing models in both central and axial electronic circular dichroism (ECD) predictions.
Methodology
The authors developed GSF-χ by integrating a Chiral-RoPE operator into a graph transformer architecture. This operator utilizes a pseudoscalar to dictate the direction of relative rotations among atoms, ensuring that the attention mechanism is invariant to reflections. The model employs a parity projection to separate even and odd outputs, allowing for consistent predictions across enantiomer pairs. The methodology includes rigorous property tests and a coordinate-reflection audit to validate the model's theoretical foundations.
Results
GSF-χ achieved state-of-the-art results on standard chirality classification benchmarks, improving axial rotation and symbol predictions by 12.6% and 7.9%, respectively, compared to the strongest baseline. The model demonstrated exact enantiomer-pair consistency with minimal accuracy trade-offs under complete supervision, and it maintained predictive capabilities even with limited mirror supervision.
Implications
The advancements presented in GSF-χ have significant implications for drug discovery and molecular design, particularly in fields where chirality plays a critical role in pharmacological efficacy and safety. The model's ability to accurately predict chiral properties could enhance the development of chiral drugs and materials, leading to more effective therapeutic options.
Groupoid-Based Internal State Representations for Reinforcement Learning with Local Symmetries
Reinforcement Learning
- Introduction of a groupoid-based framework for reinforcement learning that captures local symmetries.
- Dynamic discovery of equivalence structures allows for more efficient learning in complex environments.
- Empirical results show improved sample efficiency and convergence compared to traditional Q-learning.
- The framework supports local generalization without requiring global symmetry, enhancing robustness.
Read more
Groupoid-Based Internal State Representations for Reinforcement Learning with Local Symmetries
Summary
This paper introduces a novel framework for reinforcement learning (RL) that utilizes groupoids to capture local, state-dependent symmetries, enhancing the agent's ability to learn in complex environments. Traditional RL methods often rely on fixed group actions and predefined state abstractions, which limit their effectiveness in environments with modular and context-dependent regularities. The proposed framework allows for dynamic discovery of equivalence structures during agent-environment interactions, enabling the agent to maintain orbit representatives and transporters that map raw states to canonical forms. This approach facilitates learning and decision-making in a symmetry-reduced space while preserving local distinctions. Empirical evaluations demonstrate that the groupoid-based method significantly improves sample efficiency and convergence rates in dense and large-scale environments exhibiting partial symmetries, outperforming standard Q-learning. The findings suggest that leveraging local symmetries provides a mathematically grounded and practical pathway to scalable and generalizable reinforcement learning.
Methodology
The authors developed a reinforcement learning framework that employs groupoids to represent local symmetries. The methodology includes maintaining orbit representatives and transporters to facilitate value propagation across equivalence-related states. This enables the agent to learn in a dynamically growing canonical space, allowing for efficient experience sharing while preserving necessary distinctions.
Results
The empirical results indicate that the groupoid-based approach leads to faster convergence and lower policy variance in environments characterized by repeated structural motifs. The method demonstrates substantial performance gains over standard Q-learning, particularly in dense and large-scale settings where classical methods struggle.
Implications
The findings suggest that incorporating groupoid-based representations in reinforcement learning can significantly enhance the efficiency and robustness of learning algorithms. This approach has potential applications in various fields such as robotics, healthcare, and any domain where local symmetries can be exploited to improve decision-making processes.
Reinforcement Learning for Syndrome Extraction
Reinforcement Learning
Optimization
Theory
- Introduces FastSched, a reinforcement learning-based tool for syndrome extraction in quantum error correction.
- Achieves significant reductions in logical error rates compared to state-of-the-art tools.
- Combines quality and scalability in schedule synthesis, addressing limitations of previous methods.
- Utilizes importance sampling for efficient evaluation of schedules.
Read more
Reinforcement Learning for Syndrome Extraction
Summary
This paper addresses the challenge of syndrome extraction in quantum error correction (QEC), which is crucial for maintaining the reliability of quantum computations. The authors highlight that the number of possible syndrome extraction implementations grows exponentially with the syndrome size, leading to a search problem for finding low logical error rate (LER) implementations. Previous methods have either compromised on solution quality or scalability. The authors propose a novel approach called FastSched, which utilizes reinforcement learning (RL) combined with importance sampling to efficiently search for optimal schedules without sacrificing quality. FastSched significantly outperforms existing tools, reducing LER by 25.9% compared to AlphaSyndrome and 71.7% compared to PropHunt, achieving a remarkable 97.8% reduction for a surface code with distance 15. The methodology involves training an RL agent to learn effective gate orders and evaluate schedules using rare-event methods, which contrasts with the Monte Carlo methods used by prior tools. This work not only enhances the performance of syndrome extraction but also provides a scalable solution applicable to larger quantum codes.
Methodology
The authors developed FastSched, which employs reinforcement learning to search for optimal schedules for syndrome extraction. The RL agent learns from logical error rate estimates and selects CNOT orders based on previous choices. Importance sampling is used to evaluate schedules efficiently, focusing on a few promising candidates to reduce computational costs.
Results
FastSched outperforms existing scheduling tools, achieving a 25.9% reduction in logical error rate compared to AlphaSyndrome and a 71.7% reduction compared to PropHunt. For a surface code with distance 15, it culminates in a 97.8% reduction in logical error rate.
Implications
The findings suggest that reinforcement learning can effectively optimize quantum error correction processes, potentially leading to more reliable quantum computations. This approach could be extended to other areas of quantum computing and error correction, enhancing the scalability and efficiency of quantum algorithms.
Fed-Equilibrium Framework for Topological Pareto Control in Robust and Fair Clinical Federated Learning
Federated Learning
- Introduction of the Fed-Equilibrium framework to address knowledge dominance in clinical federated learning.
- Implementation of a two-stage gradient control cascade for enhanced robustness and fairness.
- Validation through bi-national simulations shows effective representation of minority clinical data.
- Demonstrates that minority nodes can achieve convergence similar to larger hubs, promoting equitable learning.
Read more
Fed-Equilibrium Framework for Topological Pareto Control in Robust and Fair Clinical Federated Learning
Summary
The paper introduces the Fed-Equilibrium framework, addressing the challenge of 'knowledge dominance' in Federated Learning (FL) within multi-center clinical networks. Traditional FL methods often lead to an imbalance where high-volume hubs overshadow minority community nodes, risking the loss of valuable clinical insights from smaller cohorts. Fed-Equilibrium advances from basic geometric defenses to a more sophisticated topological equilibrium approach. It employs a two-stage gradient control cascade: Stage I focuses on geometric quality assurance to filter out malicious noise, while Stage II implements topological Pareto control to identify optimal contributions from all nodes. The framework was validated through a bi-national simulation involving Canadian and U.S. clinical registries, demonstrating its ability to secure the network against adversarial influences while ensuring fair representation of underrepresented signals. The results indicate that even minority nodes can achieve convergence comparable to larger hubs, effectively countering knowledge dominance and fostering a more equitable knowledge commons in clinical federated learning.
Methodology
The Fed-Equilibrium framework combines geometric quality assurance with topological control to create a stable equilibrium in federated learning. It consists of three layers: data structuring and harmonization, a federated parameter-efficient fine-tuning network, and a synergistic gradient control cascade. The framework utilizes a cosine similarity funnel for noise filtering and a systematic parameter sweep to locate the Pareto knee point, ensuring that minority contributions are adequately represented in the global model.
Results
The experimental results from the bi-national simulation indicated that the Fed-Equilibrium framework effectively neutralizes the knowledge dominance effect, allowing minority nodes to achieve deep convergence comparable to data-rich hubs. This confirms the framework's capability to maintain both security and fairness in clinical federated learning networks.
Implications
The Fed-Equilibrium framework has significant implications for clinical federated learning, particularly in ensuring that diverse clinical patterns from smaller cohorts are preserved and utilized. This approach can enhance the generalizability of clinical models while respecting local representation, potentially improving patient outcomes and research quality in multi-center healthcare settings.
Decoding Mixture Perception through Computational Modeling of Component Interactions
Robotics
- Proposed a deep learning framework for accurate odor perception recognition of multi-molecule mixtures.
- Developed a fusion strategy that integrates multi-receptor and multi-molecule response curves.
- Achieved a recognition accuracy of 92.2% in identifying complex odor perceptions.
- Established a computational pathway from chemical blending to neural encoding and perceptual formation.
Read more
Decoding Mixture Perception through Computational Modeling of Component Interactions
Summary
This study addresses the challenge of olfactory perception, particularly in the context of multi-molecule mixtures, which are prevalent in nature. The authors propose a novel bio-inspired deep learning framework designed to accurately recognize odor perception from these complex mixtures. The framework constructs neural response curves for molecule-receptor interactions and employs a fusion strategy that integrates attention-weighted multi-receptor curves with concentration-dependent multi-molecule curves. This approach effectively simulates the competitive activation and synergistic integration of mixture components, allowing the model to transfer knowledge from single-molecule perception data to enhance recognition of mixture characteristics. The research establishes a comprehensive computational pathway from chemical blending to neural encoding and perceptual formation. The model achieved an impressive accuracy of 92.2% in recognizing odor perceptions, demonstrating its potential to advance understanding of olfactory mechanisms and improve applications in disease diagnosis and industrial safety. Furthermore, the framework can be integrated into embodied cognitive systems to enhance perceptual and interactive capabilities in complex environments.
Methodology
The authors designed a molecule-to-receptor neural response curve prediction model that characterizes interactions between molecules and receptors. They developed a fusion strategy to integrate multi-receptor and concentration-dependent multi-molecule curves, optimizing the model by minimizing discrepancies between single-molecule and multi-molecule response patterns. The model leverages existing single-molecule perception data to guide the prediction of mixture perceptions.
Results
The proposed model demonstrated a high accuracy of 92.2% in recognizing the olfactory characteristics of multi-molecule mixtures, showcasing its effectiveness in addressing the complexities of odor perception.
Implications
The findings have significant implications for advancing olfactory sensing technologies, particularly in autonomous robotic systems, enhancing their ability to assess safety and danger. The research also holds potential for applications in disease diagnosis and industrial safety monitoring.
PLSP (Pre-hoc Liminal Space Profiling): OOD Prediction over Detection -- An Anticipatory Approach for Machine Learning Model Reliability
Theory
- Introduction of the CREDibility Score (CREDS) for OOD prediction.
- Development of credibility curves and heat maps to analyze model behavior.
- Demonstration of CREDS as a valid measure of model trustworthiness.
- Ability to evaluate model performance without the need for OOD datasets.
Read more
PLSP (Pre-hoc Liminal Space Profiling): OOD Prediction over Detection -- An Anticipatory Approach for Machine Learning Model Reliability
Summary
The paper addresses the critical issue of Out-of-Distribution (OOD) data, which poses significant risks to machine learning models during deployment. Traditional OOD detection methods are post-hoc, relying on metrics like accuracy and AUC-ROC to evaluate model performance after inference. In contrast, this work introduces a pre-hoc anticipatory framework called PLSP for OOD prediction, shifting the focus from detection to prediction. The authors propose a novel dataset-independent metric, the CREDibility Score (CREDS), which assesses a model's generalization ability in the face of OOD data. Additionally, they introduce credibility curves to analyze the maximum credibility a model can achieve and credibility heat maps to characterize model behavior across various datasets. The findings indicate that CREDS provides a robust measure for enhancing model reliability against OOD data, demonstrating a strong correlation with traditional post-hoc metrics while allowing for proactive adjustments during model development.
Methodology
The authors mathematically define the CREDibility Score (CREDS) based on the Kullback-Leibler divergence between the distributions of training and OOD data features. They assess the model's response to OOD data by measuring deviations in feature distributions and aggregating these scores to produce an overall credibility score. The methodology includes the introduction of credibility curves to evaluate changes in model credibility and the use of heat maps to visualize model behavior across datasets.
Results
Experiments conducted on multiple datasets reveal that the proposed CREDS metric correlates well with traditional post-hoc accuracy metrics, demonstrating a consistent decline in both measures with increasing noise levels. This correlation validates CREDS as an effective predictor of model generalization and trustworthiness, even in the absence of OOD datasets.
Implications
The proposed framework and metrics can significantly enhance the reliability of machine learning models in real-world applications, particularly in critical areas like medical image analysis and anomaly detection, where OOD data can lead to catastrophic failures. By enabling proactive adjustments during model development, the approach can improve safety and performance in deployment.
GUIDE: Generative Utility Inference and Decision Engine
NLP
Large Language Models
Optimization
- GUIDE integrates Bayesian adaptive sampling with symbolic representation learning for effective preference elicitation.
- The architecture allows for diverse question types and produces interpretable domain-specific preference models.
- In silico experiments show GUIDE outperforms existing methods in cold-start scenarios and minimizes recommendation regret.
- The framework enhances transparency and expert oversight in the elicitation process.
Read more
GUIDE: Generative Utility Inference and Decision Engine
Summary
The paper introduces GUIDE, a novel elicitation architecture designed to measure human user preferences effectively, addressing challenges in AI alignment. Existing methods struggle with multidimensional preferences and grounding in domain knowledge. GUIDE combines Bayesian adaptive sampling for question selection with symbolic representation learning to initialize domain-specific preference models. It features an extensible type system for diverse elicitation questions and produces interpretable preference representations through symbolic rule-based learning. The architecture enhances observability and steerability, facilitating deployment and analysis of elicitation processes. Evaluated in investment portfolio optimization, GUIDE demonstrates improved cold-start performance and reduced recommendation regret compared to prior methods and LLM-only baselines, showcasing its effectiveness in early elicitation interactions across various user personas.
Methodology
GUIDE employs a Bayesian elicitation framework that utilizes dimension discovery and preference calibration processes. It incorporates an extensible set of question types for effective querying and a symbolic domain initialization pipeline to create interpretable rules and population priors, ensuring empirical grounding. The architecture is designed for transparency, enabling analysis and expert steering throughout the elicitation process.
Results
The evaluation of GUIDE in financial portfolio optimization indicates its ability to recover ground-truth preferences across diverse investor profiles. It consistently improves recommendation quality during cold-start and early-turn interactions compared to ablated versions, LLM-only approaches, and prior works such as OPEN and PEBOL.
Implications
GUIDE has significant implications for AI alignment and decision-making processes, particularly in finance. Its ability to accurately elicit and represent user preferences can enhance personalized decision support systems and reduce biases in recommendations. The framework also provides a foundation for further research into elicitation strategies and the interplay between statistical and generative AI methods.
Estimating Pedestrian Volumes from GIS-Derived Built-Environment Features: A Machine Learning Framework
Optimization
- Developed a machine learning framework for estimating pedestrian volumes from GIS data.
- Improved upon traditional Negative Binomial GLM by incorporating feature selection and gradient boosting.
- Achieved significant reductions in RMSE compared to the baseline model.
- Released code on GitHub for public access and further research.
Read more
Estimating Pedestrian Volumes from GIS-Derived Built-Environment Features: A Machine Learning Framework
Summary
This paper addresses the challenge of estimating pedestrian volumes across urban environments, which is crucial for transportation agencies to prioritize safety investments. Traditional manual counting methods are costly and limited in coverage, prompting the need for automated estimation techniques. The authors propose a machine learning framework that predicts 2-hour PM peak pedestrian volumes at 101 urban intersections in Portland, Oregon, using built-environment, land-use, and street-network features derived from open GIS data. The study builds upon a baseline Negative Binomial Generalized Linear Model (GLM) by incorporating advanced techniques such as feature selection, count-aware gradient boosting, and repeated cross-validation. The authors benchmark seven models and select the best-performing configuration based on a combined ranking of RMSE, MAPE, and SMAPE metrics. The final model, a histogram-based gradient boosting model with Poisson loss and L1 Lasso feature selection, significantly outperforms the GLM baseline, achieving a 12% reduction in cross-validated RMSE and a 19% reduction in holdout RMSE. The code for the proposed framework is made publicly available on GitHub, promoting further research and application in pedestrian volume estimation.
Methodology
The authors utilized a machine learning pipeline that includes feature selection, count-aware gradient boosting, and repeated cross-validation. They benchmarked various models and selected the best configuration based on multiple performance metrics, focusing on a dataset of 101 urban intersections with 79 candidate predictors.
Results
The selected histogram-based gradient boosting model with Poisson loss and L1 Lasso feature selection reduced the cross-validated RMSE from 89.8 to 78.7 and the holdout RMSE from 108.0 to 87.9, demonstrating a significant improvement over the baseline GLM.
Implications
The findings suggest that machine learning approaches can effectively estimate pedestrian volumes using readily available GIS data, which can aid transportation agencies in making informed decisions about infrastructure investments and safety improvements. This framework can be adapted for other urban areas, enhancing pedestrian safety and urban planning.
Information-Induced Training Geometry: Exact Reduction, Canonical Completion, and Structured Expressivity
Optimization
Theory
Efficient ML
- Establishes a framework for understanding how training data constrains optimizer geometry.
- Introduces a unique completion for full-column-rank SPD channels under Riemannian geometry.
- Derives a closed-form pair metric that separates visible-metric motion from subspace rotation.
- Characterizes exact reduction and structured expressivity in the context of finite-dimensional AIRM models.
Read more
Information-Induced Training Geometry: Exact Reduction, Canonical Completion, and Structured Expressivity
Summary
This paper investigates the geometric implications of training data on optimizer geometry through the lens of information channels. The author establishes how partial information can determine a full positive cometric relative to a reference, revealing degrees of freedom that remain unidentified. The central result is the resolution of the full-column-rank SPD channel under affine-invariant Riemannian geometry, leading to a split-Hadamard metric submetry and a unique completion that realizes a visible target. The paper elucidates how the geometry evolves as the channel moves, forming a gauge-invariant rank stratification of the SPD cone. The derived metrics separate visible-metric motion from subspace rotation, providing a framework for understanding singularities in reference-valued modes. The results also include a characterization of exact reduction, reference-dependent completion, and structured expressivity within a finite-dimensional AIRM model, contributing to the understanding of adaptive optimization methods and their geometric foundations.
Methodology
The paper employs Riemannian geometry to analyze the relationship between training data and optimizer geometry. It utilizes concepts such as metric submetries, Hadamard manifolds, and affine-invariant Riemannian metrics to derive results related to visible decision problems and their geometric properties. The author also develops closed-form expressions for metrics and completions, facilitating the analysis of rank stratifications and the behavior of optimization algorithms.
Results
The main results include the establishment of a split-Hadamard submetry for full-column-rank SPD channels, the derivation of a unique canonical completion for visible targets, and the formulation of a closed-form mismatch-weighted pair metric. These results provide insights into the geometric structure of optimization problems and the conditions for lossless visible reduction. Additionally, the paper quantifies the remaining invisible ambiguity and characterizes the structured expressivity of various optimizer families.
Implications
The findings have significant implications for the design and analysis of adaptive optimization algorithms in machine learning. By understanding the geometric constraints imposed by training data, researchers can develop more effective optimization strategies that leverage the underlying structure of the problem. This work may also inform future research on the interplay between geometry and optimization in high-dimensional spaces.
A Full Adam Theorem for Spectral Heavy-Tail Onset
Theory
Optimization
- Establishes a full Adam theorem for spectral heavy-tail onset in a closed state-evolution model.
- Derives population gradients using Stein-Hermite calculus and proves covariance concentration.
- Demonstrates the conversion of Adam momentum into a non-centered Gaussian sign kernel.
- Proves approximate-target KL contraction with upper and lower hitting bounds.
Read more
A Full Adam Theorem for Spectral Heavy-Tail Onset
Summary
This paper presents a comprehensive theorem regarding the behavior of the Adam optimization algorithm in the context of spectral heavy-tail onset within a closed Gaussian Stein-Hermite teacher-student model. The author derives the population gradient using Stein-Hermite calculus and establishes a series of results that connect Adam's momentum and denominator to a spectral hitting law. The theorem is notable for its rigorous formalization of each step in the process, demonstrating how Adam's trajectory can reach a spectral heavy-tail window after the emergence of a first spike-bulk gap. The author also discusses the impossibility of a stronger arbitrary-gradient Adam theorem and the limitations of exact two-step linear-network loss dynamics in identifying factor spectra or heavy-tail hitting times. The findings are supported by a detailed proof dependency map and various lemmas that contribute to the overall theorem.
Methodology
The methodology involves a closed Gaussian Stein-Hermite model where the author derives the Adam recurrences, applies Stein-Hermite calculus for gradient derivation, and utilizes various mathematical theorems (such as the Hermite edge-transfer theorem and Gram update) to formalize the connections between Adam's components and spectral hitting laws. The paper includes rigorous proofs and a proof dependency map to illustrate the relationships among the derived results.
Results
The main result is the establishment of a full Adam theorem that quantitatively describes how Adam's optimization trajectory reaches a spectral heavy-tail window characterized by a specific hitting time law. The theorem is formalized within the closed state-evolution model, demonstrating that the first spike-bulk spectral gap leads to this heavy-tail onset. The author also proves that arbitrary-gradient Adam theorems cannot imply heavy-tail onset and that exact two-step linear loss dynamics do not effectively identify factor spectra.
Implications
The findings have significant implications for understanding the dynamics of optimization algorithms like Adam in high-dimensional settings, particularly in relation to heavy-tail phenomena. This could influence the design of more robust optimization strategies in machine learning, especially in scenarios involving complex loss landscapes.
Robust Policy Optimization via Adversarial Importance Sampling
Reinforcement Learning
Optimization
Robotics
- Introduction of Adversarial Importance Sampling (Advis) for robust policy optimization.
- Development of advrl, a modular library for implementing and evaluating robustness methods.
- Emphasis on the need for diverse adversarial configurations in robustness evaluation.
- Demonstration of the effectiveness of the proposed method in continuous control environments.
Read more
Robust Policy Optimization via Adversarial Importance Sampling
Summary
This paper addresses the vulnerabilities of deep reinforcement learning (DRL) policies to adversarial perturbations by proposing a novel method called Adversarial Importance Sampling (Advis). The authors identify three main limitations in existing robust DRL approaches: the need for additional environment interactions, reliance on auxiliary networks, and a focus on short-term robustness. Advis overcomes these issues by utilizing importance sampling over trajectories from standard training to estimate and optimize worst-case returns without requiring extra samples or networks, thus capturing long-term robustness. Additionally, the authors introduce advrl, a modular PyTorch library designed for rapid prototyping and reproducible evaluations of robustness methods and adversarial attacks. They also highlight the importance of evaluating policies against a diverse set of adversarial configurations, demonstrating that optimal hyperparameters do not transfer across agents, which can lead to misleading assessments of robustness. The proposed method is evaluated in continuous control environments, showing consistent improvement over existing baselines.
Methodology
The authors combine convex relaxation of neural networks with importance sampling to compute a worst-case policy distribution. This allows them to estimate the discounted return under this worst-case policy using previously collected trajectories, thus optimizing for long-term robustness without additional samples or networks. The advrl library is designed to facilitate the implementation and evaluation of robustness methods in a clean and modular manner.
Results
The proposed approach consistently outperforms existing robust DRL methods in continuous control environments, demonstrating its effectiveness in enhancing policy robustness against adversarial perturbations. The evaluation involved a significantly larger set of adversarial configurations compared to prior work, leading to more reliable assessments of robustness.
Implications
The findings suggest that robust DRL can be significantly improved by addressing the limitations of existing methods, particularly in terms of evaluation practices. The advrl library can serve as a valuable tool for researchers and practitioners in the field, promoting reproducibility and ease of experimentation in developing robust reinforcement learning algorithms.