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
Rethinking Successive Rejects Algorithms This paper optimizes the predefined phase-length schedules of Successive Rejects algorithms to improve error exponents in fixed-budget best-arm identification. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Theoretical multi-armed bandit algorithm analysis, not language model evaluation.
Rethinking Successive Rejects Algorithms This paper optimizes the predefined phase-length schedules of Successive Rejects algorithms to improve error exponents in fixed-budget best-arm identification. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Studies bandit elimination schedules, not small models or efficient inference serving.
ABNAT: Attention-Based Noise-Aware Training for Robust Transformers on Analog Compute-in-Memory Systems This paper proposes an attention-based noise-aware training method to improve the robustness of Transformer models deployed on analog compute-in-memory hardware. Peer-reviewed	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on hardware noise-aware training, not LLM benchmark evaluation.
ABNAT: Attention-Based Noise-Aware Training for Robust Transformers on Analog Compute-in-Memory Systems This paper proposes an attention-based noise-aware training method to improve the robustness of Transformer models deployed on analog compute-in-memory hardware. Peer-reviewed	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Deals with analog hardware noise resilience, not safety, guardrails, or alignment.
Can AI Forecast the Extremes? Evaluating Probabilistic Weather Models on Tropical Cyclones This paper evaluates probabilistic AI weather forecasting models against operational physics-based ensembles on tropical cyclone tracking, intensity, and structure. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Evaluates weather prediction accuracy, not AI deployment safety, jailbreaks, or guardrails.
Can AI Forecast the Extremes? Evaluating Probabilistic Weather Models on Tropical Cyclones This paper evaluates probabilistic AI weather forecasting models against operational physics-based ensembles on tropical cyclone tracking, intensity, and structure. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on extreme weather forecasting models, not small or efficient language models.
Can AI Forecast the Extremes? Evaluating Probabilistic Weather Models on Tropical Cyclones This paper evaluates probabilistic AI weather forecasting models against operational physics-based ensembles on tropical cyclone tracking, intensity, and structure. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Evaluates meteorological weather models, not language models or LLM-as-a-judge.
Learning Changes the World and Then What Was Learned has Changed: Training and Deploying Neural Networks in the Control Loop This paper categorizes failure modes when deploying neural networks in closed-loop control systems and proposes online adaptation policies to counter distribution shifts. Peer-reviewed	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Concerns closed-loop control dynamics, not model compression or efficient serving.
Learning Changes the World and Then What Was Learned has Changed: Training and Deploying Neural Networks in the Control Loop This paper categorizes failure modes when deploying neural networks in closed-loop control systems and proposes online adaptation policies to counter distribution shifts. Peer-reviewed	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on control-loop dynamical error analysis, not LLM evaluation.
Scene Reconstruction Using Ubiquitous Wireless Signals via Diffusion Posterior Inference Over physics-Grounded Representations This paper presents a diffusion-based posterior inference framework to reconstruct physical scenes from ubiquitous wireless channel measurements. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on wireless channel sensing, not deployment safety, guardrails, or jailbreaks.
Scene Reconstruction Using Ubiquitous Wireless Signals via Diffusion Posterior Inference Over physics-Grounded Representations This paper presents a diffusion-based posterior inference framework to reconstruct physical scenes from ubiquitous wireless channel measurements. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on RF-based 3D scene reconstruction, not language model evaluation.
Scene Reconstruction Using Ubiquitous Wireless Signals via Diffusion Posterior Inference Over physics-Grounded Representations This paper presents a diffusion-based posterior inference framework to reconstruct physical scenes from ubiquitous wireless channel measurements. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Presents wireless signal diffusion models, not small models or efficient inference.
Beyond Fixed Acquisition Functions: Temporal Difference Active Learning for Perturbation Discovery This paper introduces a temporal difference active learning framework to adaptively learn experiment selection policies for budgeted perturbation discovery. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Addresses drug screen active learning, not model compression or efficient serving.
Beyond Fixed Acquisition Functions: Temporal Difference Active Learning for Perturbation Discovery This paper introduces a temporal difference active learning framework to adaptively learn experiment selection policies for budgeted perturbation discovery. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on biological perturbation active learning, not LLM evaluation benchmarks.
Beyond Fixed Acquisition Functions: Temporal Difference Active Learning for Perturbation Discovery This paper introduces a temporal difference active learning framework to adaptively learn experiment selection policies for budgeted perturbation discovery. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Deals with experimental biological screening, not AI deployment safety or guardrails.
The importance of tensor handling and implementation of physics-informed DeepONets Reorganizes tensor operations in physics-informed DeepONet architectures to eliminate redundancy and reduce training costs for higher-dimensional problems. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Targets scientific operator networks, not small language models or LLM efficiency.
The importance of tensor handling and implementation of physics-informed DeepONets Reorganizes tensor operations in physics-informed DeepONet architectures to eliminate redundancy and reduce training costs for higher-dimensional problems. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on physics-informed neural network architectures, not LLM benchmarks.
The importance of tensor handling and implementation of physics-informed DeepONets Reorganizes tensor operations in physics-informed DeepONet architectures to eliminate redundancy and reduce training costs for higher-dimensional problems. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on solving differential equations, not safety, guardrails, or alignment.
LCT: Losslessly Compressed Tensors with GPU friendly execution for Training Provides a GPU-accelerated Huffman compression framework for tensors to lower memory overhead during model training. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Deals with training memory compression, not AI deployment safety or guardrails.
LCT: Losslessly Compressed Tensors with GPU friendly execution for Training Provides a GPU-accelerated Huffman compression framework for tensors to lower memory overhead during model training. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on GPU tensor compression systems, not LLM benchmark evaluation.
Attention over noisy examples: generalization and memorization in the exponential Hopfield network Analyzes generalization and memorization dynamics in exponential Hopfield networks trained with softmax attention over noisy examples. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Analyzes theoretical Hopfield models, not practical model compression or efficiency.
Attention over noisy examples: generalization and memorization in the exponential Hopfield network Analyzes generalization and memorization dynamics in exponential Hopfield networks trained with softmax attention over noisy examples. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Studies associative memory theory rather than AI deployment safety or guardrails.
Attention over noisy examples: generalization and memorization in the exponential Hopfield network Analyzes generalization and memorization dynamics in exponential Hopfield networks trained with softmax attention over noisy examples. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Statistical physics analysis of Hopfield networks, not LLM evaluation benchmarks.
Generalization Guarantees of Randomized Block Coordinate Descent for Nonconvex Learning Derives theoretical high-probability generalization guarantees and early stopping bounds for randomized block coordinate descent on nonconvex problems. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Theoretical optimization theory, not deployment safety, guardrails, or hallucination.
Generalization Guarantees of Randomized Block Coordinate Descent for Nonconvex Learning Derives theoretical high-probability generalization guarantees and early stopping bounds for randomized block coordinate descent on nonconvex problems. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Theoretical nonconvex optimization analysis, not LLM evaluation or benchmarking.
Generalization Guarantees of Randomized Block Coordinate Descent for Nonconvex Learning Derives theoretical high-probability generalization guarantees and early stopping bounds for randomized block coordinate descent on nonconvex problems. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Presents theoretical generalization bounds, not small models or efficient inference.
Architecting for Hardware Diversity: Meta's End-to-End Co-Design for High-Performance AI Training on AMD GPUs Shares hardware-model co-design techniques and analytical kernel selection for training large recommendation models on AMD GPUs. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on hardware-model co-design, not LLM evaluation or benchmark methodology.
Architecting for Hardware Diversity: Meta's End-to-End Co-Design for High-Performance AI Training on AMD GPUs Shares hardware-model co-design techniques and analytical kernel selection for training large recommendation models on AMD GPUs. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Addresses hardware optimization and porting, not AI safety, hallucination, or guardrails.
Query-Efficient Rejection Sampling for Parallel Masked Diffusion Proposes a single-increment rejection sampling algorithm to correct token correlation bias efficiently in parallel masked diffusion. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on diffusion sampling algorithms, not language model evaluation benchmarks.
Query-Efficient Rejection Sampling for Parallel Masked Diffusion Proposes a single-increment rejection sampling algorithm to correct token correlation bias efficiently in parallel masked diffusion. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Presents diffusion sampling theory, not AI guardrails, jailbreaks, or safety.
SurvICL: a Foundation Model for in-context Survival Analysis Introduces a tabular foundation model pretrained on synthetic priors to predict survival curves in-context. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Studies clinical survival predictions rather than AI safety or reliability guardrails.
SurvICL: a Foundation Model for in-context Survival Analysis Introduces a tabular foundation model pretrained on synthetic priors to predict survival curves in-context. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Focuses on in-context survival models, not model compression or efficient serving.
SurvICL: a Foundation Model for in-context Survival Analysis Introduces a tabular foundation model pretrained on synthetic priors to predict survival curves in-context. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Introduces a tabular survival analysis benchmark, not LLM or language model evaluation.
Explore Once, Evaluate Many: Multiple Policy Evaluation for Stochastic Linear Bandits Introduces an algorithm to evaluate multiple target policies simultaneously in contextual linear bandits with shared sampling. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Studies stochastic bandit sample complexity, not efficient or small model inference.
Explore Once, Evaluate Many: Multiple Policy Evaluation for Stochastic Linear Bandits Introduces an algorithm to evaluate multiple target policies simultaneously in contextual linear bandits with shared sampling. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Theoretical bandit policy evaluation, not benchmarking or evaluating LLMs.
Explore Once, Evaluate Many: Multiple Policy Evaluation for Stochastic Linear Bandits Introduces an algorithm to evaluate multiple target policies simultaneously in contextual linear bandits with shared sampling. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on bandit sample complexity bounds, not deployment safety or jailbreaks.
BitSparse: Compact Activation Storage for ReLU Feed-Forward Networks Presents a GPU format that losslessly compresses sparse activations to reduce memory usage during neural network training. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on memory compression engineering, not evaluating or benchmarking LLMs.
BitSparse: Compact Activation Storage for ReLU Feed-Forward Networks Presents a GPU format that losslessly compresses sparse activations to reduce memory usage during neural network training. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on memory reduction techniques, not safety, guardrails, or alignment.
Topology-Inspired Fine-Tuning for Wind-Field Super-Resolution Applies topology-inspired loss functions to fine-tune super-resolution neural networks on scientific wind-field datasets. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Presents scientific image super-resolution, not efficient or small language models.
Topology-Inspired Fine-Tuning for Wind-Field Super-Resolution Applies topology-inspired loss functions to fine-tune super-resolution neural networks on scientific wind-field datasets. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on weather super-resolution fine-tuning, not AI safety and reliability.
Topology-Inspired Fine-Tuning for Wind-Field Super-Resolution Applies topology-inspired loss functions to fine-tune super-resolution neural networks on scientific wind-field datasets. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Evaluates wind-field super-resolution topology, not language model benchmarking.
Baselines and Score Centering in Policy Gradient Estimation Analyzes score centering and covariance estimands to maintain reward shift invariance under regularized importance sampling. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Covers policy gradient math for post-training, not model compression or efficiency.
Baselines and Score Centering in Policy Gradient Estimation Analyzes score centering and covariance estimands to maintain reward shift invariance under regularized importance sampling. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Theoretical analysis of policy gradient estimation, not LLM evaluation benchmarks.
Baselines and Score Centering in Policy Gradient Estimation Analyzes score centering and covariance estimands to maintain reward shift invariance under regularized importance sampling. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Discusses RL post-training optimization theory rather than deployment safety or guardrails.
Learning Rational Strategies for Temporal Objectives in Multi-Agent Systems Analyzes rational strategy synthesis and equilibrium guarantees for stochastic multi-agent systems with probabilistic temporal logic objectives. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not study efficient serving, quantization, or small language models.
Learning Rational Strategies for Temporal Objectives in Multi-Agent Systems Analyzes rational strategy synthesis and equilibrium guarantees for stochastic multi-agent systems with probabilistic temporal logic objectives. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Presents theoretical RL verification rather than testing or benchmarking LLMs.
Learning Rational Strategies for Temporal Objectives in Multi-Agent Systems Analyzes rational strategy synthesis and equilibrium guarantees for stochastic multi-agent systems with probabilistic temporal logic objectives. Preprint	AI agents via arxiv · 17 Oct, 06:18	Not relevant Theoretical multi-agent RL synthesis, not language model agents or planning.
Learning Rational Strategies for Temporal Objectives in Multi-Agent Systems Analyzes rational strategy synthesis and equilibrium guarantees for stochastic multi-agent systems with probabilistic temporal logic objectives. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on game-theoretic synthesis rather than AI safety guardrails or hallucinations.
Backward Compatibility in Zero-Bit Neural Audio Watermarking Establishes a training approach to ensure backward compatibility and prevent false positives across versions of neural audio watermarking. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not involve model pruning, distillation, quantization, or efficient serving.
Backward Compatibility in Zero-Bit Neural Audio Watermarking Establishes a training approach to ensure backward compatibility and prevent false positives across versions of neural audio watermarking. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Covers audio watermarking backward compatibility rather than LLM evaluation benchmarks.
Backward Compatibility in Zero-Bit Neural Audio Watermarking Establishes a training approach to ensure backward compatibility and prevent false positives across versions of neural audio watermarking. Preprint	Multimodal models via arxiv · 17 Oct, 06:18	Not relevant Focuses on zero-bit audio watermarking, not multimodal reasoning or vision-language models.
Backward Compatibility in Zero-Bit Neural Audio Watermarking Establishes a training approach to ensure backward compatibility and prevent false positives across versions of neural audio watermarking. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on audio watermarking compatibility instead of AI guardrails or jailbreaks.
Robust Reinforcement Learning for Dynamic Treatment Regimes under Model and Measurement Uncertainty Develops a robust reinforcement learning framework for dynamic treatment policies under patient measurement and model uncertainty. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Concerns clinical RL robustness rather than AI deployment safety or guardrails.
Robust Reinforcement Learning for Dynamic Treatment Regimes under Model and Measurement Uncertainty Develops a robust reinforcement learning framework for dynamic treatment policies under patient measurement and model uncertainty. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Does not address model compression, quantization, or efficient inference techniques.
Robust Reinforcement Learning for Dynamic Treatment Regimes under Model and Measurement Uncertainty Develops a robust reinforcement learning framework for dynamic treatment policies under patient measurement and model uncertainty. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on medical RL treatment regimes, not evaluating language models.
Lean Graph Search for Formal Theorem Proving Introduces a graph-based retrieval method that traces multi-hop library dependencies to find lemmas for formal theorem proving. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Addresses premise retrieval algorithms, not model quantization or efficient inference.
Lean Graph Search for Formal Theorem Proving Introduces a graph-based retrieval method that traces multi-hop library dependencies to find lemmas for formal theorem proving. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:18	Not relevant Focuses on lemma retrieval for theorem provers, not LLM evaluation benchmarks.
Lean Graph Search for Formal Theorem Proving Introduces a graph-based retrieval method that traces multi-hop library dependencies to find lemmas for formal theorem proving. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Focuses on mathematical lemma retrieval, not AI safety or deployment robustness.
PathFM: Fast Generative Forecasting of Multivariate Trajectories Uses flow matching and distillation to generate full multivariate time-series trajectory forecasts rapidly. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:18	Not relevant Deals with time series forecasting calibration rather than LLM safety or deployment guardrails.
PathFM: Fast Generative Forecasting of Multivariate Trajectories Uses flow matching and distillation to generate full multivariate time-series trajectory forecasts rapidly. Preprint	Small and efficient models via arxiv · 17 Oct, 06:18	Not relevant Distills time series trajectory generators rather than small language models.

4,021–4,080 of 17,313← Previous Next →