Every paper checked — nothing hidden

All papers

Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.

All17,313 In brief24 More (on-topic)3,333 Not relevant13,754 Not assessed87 Older352 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
Compatibility Matters: A Framework for Token-Level Teacher Selection in On-Policy Distillation The authors propose a theoretical and empirical framework to route token-level teacher selection during on-policy student distillation. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Introduces a distillation framework rather than benchmark datasets or evaluation methodologies.
Rethinking Learning from Label Proportions via Moment Matching The paper presents a moment-matching framework for weakly supervised learning when labels are grouped into bags non-randomly. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Weakly supervised classification theory, not small model distillation or efficient inference.
Rethinking Learning from Label Proportions via Moment Matching The paper presents a moment-matching framework for weakly supervised learning when labels are grouped into bags non-randomly. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on label-proportion statistical learning, not deployment safety guardrails or robustness.
Rethinking Learning from Label Proportions via Moment Matching The paper presents a moment-matching framework for weakly supervised learning when labels are grouped into bags non-randomly. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Addresses learning from label proportions, not LLM benchmark reliability or evaluation.
Regret Lower Bound for Online Learning of Quantum States with Logarithmic Loss, with Applications to Quantum State Tomography and Confidence Sequences for Matrix Means The authors establish regret lower bounds for online quantum state learning under logarithmic loss with applications to tomography. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on theoretical quantum state learning regret bounds, not LLM benchmarks.
Regret Lower Bound for Online Learning of Quantum States with Logarithmic Loss, with Applications to Quantum State Tomography and Confidence Sequences for Matrix Means The authors establish regret lower bounds for online quantum state learning under logarithmic loss with applications to tomography. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Mathematical quantum learning paper, not model compression or efficient LLM inference.
Regret Lower Bound for Online Learning of Quantum States with Logarithmic Loss, with Applications to Quantum State Tomography and Confidence Sequences for Matrix Means The authors establish regret lower bounds for online quantum state learning under logarithmic loss with applications to tomography. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Deals with quantum tomography regret bounds, not AI deployment safety or jailbreaks.
Can Capability Enhancement be Localized? The study investigates whether removing small subsets of attention heads can improve language model performance on specific tasks. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on capability improvement via head removal, not safety guardrails or robustness.
Can Capability Enhancement be Localized? The study investigates whether removing small subsets of attention heads can improve language model performance on specific tasks. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Examines head pruning mechanics rather than benchmarking or evaluation methodology.
Personalized Projected Ascent for Federated Stackelberg Optimization The paper analyzes personalized projected ascent algorithms for federated minimax problems with client-specific maximization variables. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Theoretical optimization paper, not LLM safety, guardrails, or deployment reliability.
Personalized Projected Ascent for Federated Stackelberg Optimization The paper analyzes personalized projected ascent algorithms for federated minimax problems with client-specific maximization variables. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on federated optimization theory, not small model efficiency or distillation.
Personalized Projected Ascent for Federated Stackelberg Optimization The paper analyzes personalized projected ascent algorithms for federated minimax problems with client-specific maximization variables. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Analyzes federated minimax optimization rates, not LLM benchmark or evaluation methodology.
Spectrum Is Not Enough: A Random-Matrix Account of Data-Free Preconditioning for Differentially Private Learning The authors use random matrix theory to establish theoretical limits on synthetic data-free preconditioning in differentially private learning. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Deals with differential privacy preconditioning, not model compression or efficient serving.
Spectrum Is Not Enough: A Random-Matrix Account of Data-Free Preconditioning for Differentially Private Learning The authors use random matrix theory to establish theoretical limits on synthetic data-free preconditioning in differentially private learning. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Theoretical analysis of private learning preconditioning, not deployment safety or reliability.
Spectrum Is Not Enough: A Random-Matrix Account of Data-Free Preconditioning for Differentially Private Learning The authors use random matrix theory to establish theoretical limits on synthetic data-free preconditioning in differentially private learning. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on theoretical random-matrix bounds, not LLM benchmark evaluation.
Dimension-Free PAC-Bayes Generalization for Deep Neural Networks via Dirichlet Processes The paper develops dimension-free PAC-Bayesian generalization bounds for deep neural networks using Dirichlet processes over data distributions. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Investigates mathematical generalization bounds rather than efficient model compression.
Dimension-Free PAC-Bayes Generalization for Deep Neural Networks via Dirichlet Processes The paper develops dimension-free PAC-Bayesian generalization bounds for deep neural networks using Dirichlet processes over data distributions. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Theoretical PAC-Bayes generalization paper, not an LLM benchmark.
Dimension-Free PAC-Bayes Generalization for Deep Neural Networks via Dirichlet Processes The paper develops dimension-free PAC-Bayesian generalization bounds for deep neural networks using Dirichlet processes over data distributions. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Theoretical generalization analysis without AI deployment safety focus.
Online Quantile Omniprediction for Proper Losses The authors introduce online algorithms to generate quantile forecasts that remain accurate across multiple proper loss functions simultaneously. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on loss calibration theory, not efficient model serving.
Online Quantile Omniprediction for Proper Losses The authors introduce online algorithms to generate quantile forecasts that remain accurate across multiple proper loss functions simultaneously. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Statistical quantile omniprediction theory, not an LLM benchmark.
Online Quantile Omniprediction for Proper Losses The authors introduce online algorithms to generate quantile forecasts that remain accurate across multiple proper loss functions simultaneously. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Theoretical quantile forecasting analysis, not AI system safety.
Honesty is the Best Policy ... or is it? Demand Management for Probabilistic Service Options The paper models demand management using probabilistic service options and analyzes provider incentives under transparency constraints. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Deals with operations research demand management, not small models.
Honesty is the Best Policy ... or is it? Demand Management for Probabilistic Service Options The paper models demand management using probabilistic service options and analyzes provider incentives under transparency constraints. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on customer choice manipulation in operations research, not AI safety.
Honesty is the Best Policy ... or is it? Demand Management for Probabilistic Service Options The paper models demand management using probabilistic service options and analyzes provider incentives under transparency constraints. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Studies revenue management decision processes, not LLM evaluation.
Two Sample Sizes in Test-Time LLM Ensembling: Buffered ERM and Minimax Theory The authors analyze theoretical sample-complexity tradeoffs between questions and generations when ensembling multiple language models at test time. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Theoretical statistical analysis of ensembling without model compression focus.
Two Sample Sizes in Test-Time LLM Ensembling: Buffered ERM and Minimax Theory The authors analyze theoretical sample-complexity tradeoffs between questions and generations when ensembling multiple language models at test time. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Mathematical analysis of ensembling risk rather than deployment safety.
Same Target, Different Origins: Aligning Latent Rollout for Consistent PDE Emulation The paper introduces a regularization technique that aligns internal representations across rollout origins to improve neural differential equation emulation. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Studies PDE latent consistency rather than small model efficiency.
Same Target, Different Origins: Aligning Latent Rollout for Consistent PDE Emulation The paper introduces a regularization technique that aligns internal representations across rollout origins to improve neural differential equation emulation. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on scientific neural PDE solvers, not LLM evaluation.
Same Target, Different Origins: Aligning Latent Rollout for Consistent PDE Emulation The paper introduces a regularization technique that aligns internal representations across rollout origins to improve neural differential equation emulation. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Deals with scientific neural PDE emulation, not AI safety.
Does Compression Imply Generalization? A Minimum Description Length Perspective The authors show how prequential regret from minimum description length bounds neural network generalization error without held-out data. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Theoretical learning theory on generalization bounds, not an LLM benchmark.
Does Compression Imply Generalization? A Minimum Description Length Perspective The authors show how prequential regret from minimum description length bounds neural network generalization error without held-out data. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Studies MDL theory bounds rather than practical model compression or efficiency.
Does Compression Imply Generalization? A Minimum Description Length Perspective The authors show how prequential regret from minimum description length bounds neural network generalization error without held-out data. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Covers theoretical machine learning bounds, not deployment safety.
RDL-Bench: A Benchmark for Task-Aware Question Answering over Relational Databases The paper introduces a benchmark to evaluate question answering systems on reasoning and predictive tasks over relational databases. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Proposes a database QA benchmark without model efficiency contributions.
RDL-Bench: A Benchmark for Task-Aware Question Answering over Relational Databases The paper introduces a benchmark to evaluate question answering systems on reasoning and predictive tasks over relational databases. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Benchmark paper for database QA, not focused on AI safety.
Fault Detection and Bit-Error Rate Diagnostics in Optical Wireless Links Using Edge-AI The authors implement an on-device machine learning diagnostic engine on a microcontroller for optical wireless communication links. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Evaluates optical wireless telemetry diagnostics, not LLM benchmarks.
Fault Detection and Bit-Error Rate Diagnostics in Optical Wireless Links Using Edge-AI The authors implement an on-device machine learning diagnostic engine on a microcontroller for optical wireless communication links. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on telecommunication link diagnostics rather than AI system safety.
Learning to Forecast on Multiple Spatiotemporal Hierarchies The paper proposes a graph neural network framework to exploit multiple spatiotemporal aggregation hierarchies for time-series forecasting. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Proposes graph deep learning for forecasting, not model compression.
Learning to Forecast on Multiple Spatiotemporal Hierarchies The paper proposes a graph neural network framework to exploit multiple spatiotemporal aggregation hierarchies for time-series forecasting. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses time-series aggregation architectures, not AI deployment safety.
Learning to Forecast on Multiple Spatiotemporal Hierarchies The paper proposes a graph neural network framework to exploit multiple spatiotemporal aggregation hierarchies for time-series forecasting. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on spatiotemporal time-series forecasting, not LLM benchmarking.
What Does History Add? Calendar Controls and Information-Timing Audits for University Trajectory Forecasting This study audits how historical information and calendar controls affect trajectory forecasting within a university simulator. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Investigates statistical simulator forecasting, not efficient model deployment.
What Does History Add? Calendar Controls and Information-Timing Audits for University Trajectory Forecasting This study audits how historical information and calendar controls affect trajectory forecasting within a university simulator. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Studies trajectory simulator predictions, not AI deployment safety.
What Does History Add? Calendar Controls and Information-Timing Audits for University Trajectory Forecasting This study audits how historical information and calendar controls affect trajectory forecasting within a university simulator. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Audits university simulator forecasting rather than evaluating language models.
Policy-Aligned Predictive Verification for Adaptive VLA Action-Chunk Execution The paper presents a predictive verifier that decides how many actions from a vision-language-action chunk to execute safely. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Proposes an execution verifier for robot actions, not an LLM benchmark.
Policy-Aligned Predictive Verification for Adaptive VLA Action-Chunk Execution The paper presents a predictive verifier that decides how many actions from a vision-language-action chunk to execute safely. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Targets robotic manipulation task success rather than AI deployment safety risks.
Policy-Aligned Predictive Verification for Adaptive VLA Action-Chunk Execution The paper presents a predictive verifier that decides how many actions from a vision-language-action chunk to execute safely. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:18	Not relevant Focuses on general robotic manipulation policies, not industrial defect inspection.
Policy-Aligned Predictive Verification for Adaptive VLA Action-Chunk Execution The paper presents a predictive verifier that decides how many actions from a vision-language-action chunk to execute safely. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on action chunk selection rather than model compression.
Each Reward to Its Relevant Tokens: Reward-Specific Token-Level Credit Assignment for Multi-Reward RL The authors propose an algorithm that assigns multi-reward reinforcement learning credit specifically to relevant response tokens. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on multi-reward optimization for instruction following rather than deployment safety guardrails.
Each Reward to Its Relevant Tokens: Reward-Specific Token-Level Credit Assignment for Multi-Reward RL The authors propose an algorithm that assigns multi-reward reinforcement learning credit specifically to relevant response tokens. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Proposes a multi-reward RL training method, not an LLM benchmark.
Each Reward to Its Relevant Tokens: Reward-Specific Token-Level Credit Assignment for Multi-Reward RL The authors propose an algorithm that assigns multi-reward reinforcement learning credit specifically to relevant response tokens. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on reinforcement learning credit assignment rather than efficient inference.
DuET: Note-Frame Dual Encoding for Piano Performance-to-Score Transcription The paper introduces a dual-encoder transformer that jointly transcribes piano performances to scores and tracks metrical beat frames. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Music score transcription does not address AI deployment safety.
DuET: Note-Frame Dual Encoding for Piano Performance-to-Score Transcription The paper introduces a dual-encoder transformer that jointly transcribes piano performances to scores and tracks metrical beat frames. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on music transcription evaluation, not LLM benchmarks.
DuET: Note-Frame Dual Encoding for Piano Performance-to-Score Transcription The paper introduces a dual-encoder transformer that jointly transcribes piano performances to scores and tracks metrical beat frames. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Presents a dual transformer for piano transcription without efficiency focus.
Kolmogorov-Arnold Networks: Curse of Dimensionality and Structural Bias This paper analyzes Kolmogorov-Arnold networks to show that favorable dimension scaling depends on target smoothness rather than the architecture alone. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Theoretical network capacity analysis without AI safety focus.
Kolmogorov-Arnold Networks: Curse of Dimensionality and Structural Bias This paper analyzes Kolmogorov-Arnold networks to show that favorable dimension scaling depends on target smoothness rather than the architecture alone. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Theoretical study of Kolmogorov-Arnold networks, not LLM benchmark evaluation.
Kolmogorov-Arnold Networks: Curse of Dimensionality and Structural Bias This paper analyzes Kolmogorov-Arnold networks to show that favorable dimension scaling depends on target smoothness rather than the architecture alone. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Analyzes mathematical approximation bounds rather than practical model compression methods.
Phase Kernel Lifts Capacity of Dense Associative Memory The authors introduce a nonlinear phase kernel to expand the effective retrieval capacity of modern Hopfield networks. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Studies associative memory kernels rather than LLM benchmark datasets.
Phase Kernel Lifts Capacity of Dense Associative Memory The authors introduce a nonlinear phase kernel to expand the effective retrieval capacity of modern Hopfield networks. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Examines memory network retrieval capacity rather than deployment safety.
Phase Kernel Lifts Capacity of Dense Associative Memory The authors introduce a nonlinear phase kernel to expand the effective retrieval capacity of modern Hopfield networks. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on theoretical Hopfield network capacity, not efficient model serving.
When the KV Cache Is Slower Than the GPU: Scaling Remote KV Tiers for Context-Parallel Prefill The paper analyzes performance bottlenecks when using remote key-value caches during context-parallel prefill in large language model serving. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on serving infrastructure performance rather than LLM evaluation benchmarks.
When the KV Cache Is Slower Than the GPU: Scaling Remote KV Tiers for Context-Parallel Prefill The paper analyzes performance bottlenecks when using remote key-value caches during context-parallel prefill in large language model serving. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses latency and systems throughput, not AI safety risks.

3,841–3,900 of 17,313← Previous Next →