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
Solvating machine-learning interatomic potentials for solution-phase chemistry This paper introduces an implicit solvation potential providing solvent-dependent corrections to gas-phase machine learning interatomic potentials without retraining. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on chemical potential modeling rather than efficient model inference.
Solvating machine-learning interatomic potentials for solution-phase chemistry This paper introduces an implicit solvation potential providing solvent-dependent corrections to gas-phase machine learning interatomic potentials without retraining. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Develops interatomic potentials for chemistry, not LLM evaluation or benchmarks.
MIRRA: Fast, High-Fidelity Molecular Generation with Chemistry-Informed Discrete Flow Matching This paper presents a discrete flow-matching generator that evaluates molecular identity probabilities on demand to accelerate high-fidelity molecular generation. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Presents a molecular flow-matching generative model, not LLM benchmarks.
MIRRA: Fast, High-Fidelity Molecular Generation with Chemistry-Informed Discrete Flow Matching This paper presents a discrete flow-matching generator that evaluates molecular identity probabilities on demand to accelerate high-fidelity molecular generation. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with molecular generation rather than safety or reliability guardrails.
MIRRA: Fast, High-Fidelity Molecular Generation with Chemistry-Informed Discrete Flow Matching This paper presents a discrete flow-matching generator that evaluates molecular identity probabilities on demand to accelerate high-fidelity molecular generation. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on chemistry-informed molecular generation speed, not efficient language models.
LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators This paper presents a distillation method using frozen language models to guide molecular flow generators toward natural language discovery goals. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Introduces a distillation method for molecular generators, not evaluation benchmarks.
LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators This paper presents a distillation method using frozen language models to guide molecular flow generators toward natural language discovery goals. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Applies distillation to scientific discovery rather than deployment safety or guardrails.
Value-Preserving Action Quantization The authors propose an action quantization method that preserves value functions by minimizing reachability error for search-based robot policy learning. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on continuous action space discretization in RL, not model parameter quantization.
Value-Preserving Action Quantization The authors propose an action quantization method that preserves value functions by minimizing reachability error for search-based robot policy learning. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:18	Not relevant Deals with reinforcement learning action quantization for robotics, not industrial computer vision.
Value-Preserving Action Quantization The authors propose an action quantization method that preserves value functions by minimizing reachability error for search-based robot policy learning. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Introduces robot action quantization for MCTS, not LLM benchmarks.
Value-Preserving Action Quantization The authors propose an action quantization method that preserves value functions by minimizing reachability error for search-based robot policy learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on robot trajectory planning rather than deployment safety guardrails.
Protocol Choices Matter More Than the Aggregation Algorithm: Federated Learning for In-Hospital Mortality Prediction in Care-Unit and Multi-Hospital Federations The authors compare federated learning algorithms for hospital mortality prediction and report that evaluation protocol choices outweigh aggregation algorithm differences. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Evaluates tabular federated learning protocols for hospital mortality, not LLMs.
Protocol Choices Matter More Than the Aggregation Algorithm: Federated Learning for In-Hospital Mortality Prediction in Care-Unit and Multi-Hospital Federations The authors compare federated learning algorithms for hospital mortality prediction and report that evaluation protocol choices outweigh aggregation algorithm differences. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on medical federated learning protocols rather than LLM safety or reliability.
Protocol Choices Matter More Than the Aggregation Algorithm: Federated Learning for In-Hospital Mortality Prediction in Care-Unit and Multi-Hospital Federations The authors compare federated learning algorithms for hospital mortality prediction and report that evaluation protocol choices outweigh aggregation algorithm differences. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies federated learning aggregation protocols, not efficient model compression or serving.
Colored Graph Signal Diffusion Models The authors propose a score-based diffusion framework that controls generation speed across different graph frequency bands via colored noise priors. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes graph diffusion processes, not LLM benchmarks or evaluation.
Colored Graph Signal Diffusion Models The authors propose a score-based diffusion framework that controls generation speed across different graph frequency bands via colored noise priors. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies graph signal diffusion, not small or efficient language models.
Colored Graph Signal Diffusion Models The authors propose a score-based diffusion framework that controls generation speed across different graph frequency bands via colored noise priors. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Presents graph diffusion dynamics, not deployment safety or reliability.
The Surprising Expressivity of a Single Neuron with Autoregressive Chain-of-Thought The authors demonstrate that a single autoregressive neuron using chain-of-thought generation can simulate any feedforward circuit of the same activation type. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Theoretical expressivity bounds for single neurons, not practical model efficiency.
The Surprising Expressivity of a Single Neuron with Autoregressive Chain-of-Thought The authors demonstrate that a single autoregressive neuron using chain-of-thought generation can simulate any feedforward circuit of the same activation type. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on circuit simulation theory rather than deployment safety or robustness.
The Surprising Expressivity of a Single Neuron with Autoregressive Chain-of-Thought The authors demonstrate that a single autoregressive neuron using chain-of-thought generation can simulate any feedforward circuit of the same activation type. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical analysis of single-neuron expressivity, not LLM evaluation or benchmarks.
Which Geometry on Which Layer? Measured Sign-Spectral Assignment in Mixed-Optimizer Training The authors provide a theoretical criterion to decide whether neural network layers should receive sign-based or spectral optimization updates during training. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on optimization theory rather than deployment safety, guardrails, or hallucination.
Which Geometry on Which Layer? Measured Sign-Spectral Assignment in Mixed-Optimizer Training The authors provide a theoretical criterion to decide whether neural network layers should receive sign-based or spectral optimization updates during training. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Investigates optimization layer geometry rather than LLM benchmarks or evaluation methods.
Which Geometry on Which Layer? Measured Sign-Spectral Assignment in Mixed-Optimizer Training The authors provide a theoretical criterion to decide whether neural network layers should receive sign-based or spectral optimization updates during training. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies general neural network optimizer geometry, not model efficiency or compression.
Calibrating LLM Explanations to Human Decisions with Reinforcement Learning The authors introduce a reinforcement learning method that optimizes language model explanations specifically to improve human downstream decision accuracy. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Does not address model compression, quantization, distillation, or serving efficiency.
Calibrating LLM Explanations to Human Decisions with Reinforcement Learning The authors introduce a reinforcement learning method that optimizes language model explanations specifically to improve human downstream decision accuracy. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on RL training for explanations rather than evaluation methods.
A Geometrical Approach to Constructing Neural Networks for Perfect Training-set Classification The authors develop a geometric construction method to build two-hidden-layer neural networks that achieve exact classification on given training sets. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on geometric construction of toy classifiers, not LLM benchmarks or evaluation.
A Geometrical Approach to Constructing Neural Networks for Perfect Training-set Classification The authors develop a geometric construction method to build two-hidden-layer neural networks that achieve exact classification on given training sets. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Discusses exact geometric hyperplane fitting for classification, not small LLM architectures or efficiency.
A Geometrical Approach to Constructing Neural Networks for Perfect Training-set Classification The authors develop a geometric construction method to build two-hidden-layer neural networks that achieve exact classification on given training sets. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Presents a training-set overfitting method rather than deployment safety, robustness, or reliability.
Deeply unrolled alternating projections for phase retrieval The authors propose a deep unrolling algorithm combining alternating projections with a learned denoiser for phase retrieval under changing measurement conditions. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on signal reconstruction techniques, not small or efficient language models.
Deeply unrolled alternating projections for phase retrieval The authors propose a deep unrolling algorithm combining alternating projections with a learned denoiser for phase retrieval under changing measurement conditions. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Covers inverse phase retrieval algorithms, not language model benchmarking.
Deeply unrolled alternating projections for phase retrieval The authors propose a deep unrolling algorithm combining alternating projections with a learned denoiser for phase retrieval under changing measurement conditions. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Investigates inverse imaging algorithms rather than AI safety, guardrails, or alignment.
Near-Optimal Regret Bounds for Monotone Stochastic Optimization under Semi-Bandit Feedback The authors derive an online-to-offline reduction achieving near-optimal regret bounds for strongly monotone stochastic optimization under semi-bandit feedback. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on bandit regret bounds, not small models or efficient inference techniques.
Near-Optimal Regret Bounds for Monotone Stochastic Optimization under Semi-Bandit Feedback The authors derive an online-to-offline reduction achieving near-optimal regret bounds for strongly monotone stochastic optimization under semi-bandit feedback. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses theoretical multi-armed bandit optimization without relation to AI deployment safety.
Near-Optimal Regret Bounds for Monotone Stochastic Optimization under Semi-Bandit Feedback The authors derive an online-to-offline reduction achieving near-optimal regret bounds for strongly monotone stochastic optimization under semi-bandit feedback. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical analysis of stochastic semi-bandits, not LLM evaluation or benchmarking.
GPU Goldmines: Tunable WebGPU Kernels for Edge LLM Inference The authors present an autotuning framework for WebGPU kernels to improve cross-platform inference speeds for language models on edge devices. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Presents GPU kernel implementations rather than LLM benchmarks or evaluation methods.
GPU Goldmines: Tunable WebGPU Kernels for Edge LLM Inference The authors present an autotuning framework for WebGPU kernels to improve cross-platform inference speeds for language models on edge devices. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on WebGPU execution performance rather than AI safety, red teaming, or alignment.
Accounting for model error in ensemble variational data assimilation through learned generative dynamics The authors present an ensemble variational data assimilation algorithm that incorporates learned generative forecasting models to account for complex model errors. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Covers geophysical data assimilation, not language model evaluation.
Accounting for model error in ensemble variational data assimilation through learned generative dynamics The authors present an ensemble variational data assimilation algorithm that incorporates learned generative forecasting models to account for complex model errors. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Deals with weather and climate forecast models, not small language models or efficiency.
Accounting for model error in ensemble variational data assimilation through learned generative dynamics The authors present an ensemble variational data assimilation algorithm that incorporates learned generative forecasting models to account for complex model errors. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on physical data assimilation error rather than AI safety or reliability guardrails.
Monitoring for Shifts at Unknown Times: Parameter-Free Conditional Conformal Martingales The authors introduce a parameter-free conformal martingale method modeled as a universal portfolio to quickly detect distribution shifts at unknown times. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Develops conformal martingales for distribution shift, not LLM benchmarks or evaluations.
Monitoring for Shifts at Unknown Times: Parameter-Free Conditional Conformal Martingales The authors introduce a parameter-free conformal martingale method modeled as a universal portfolio to quickly detect distribution shifts at unknown times. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on sequential shift detection algorithms, not model compression or efficient serving.
HarnessTTT: Learn to Optimize Harness at Test-Time for Scientific Discovery The authors propose a test-time learning method that trains a proposer to generate prompts, tools, and controls guiding scientific discovery agents. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on discovery performance optimization rather than safety guardrails, jailbreaks, or robustness.
HarnessTTT: Learn to Optimize Harness at Test-Time for Scientific Discovery The authors propose a test-time learning method that trains a proposer to generate prompts, tools, and controls guiding scientific discovery agents. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on test-time harness optimization for scientific discovery, not benchmark evaluation design.
HarnessTTT: Learn to Optimize Harness at Test-Time for Scientific Discovery The authors propose a test-time learning method that trains a proposer to generate prompts, tools, and controls guiding scientific discovery agents. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses prompt and tool optimization for discovery, not model quantization, distillation, or serving.
Robustly Reliable Learning for Regression The authors analyze robustly reliable learning for regression under data corruptions and propose an algorithm to compute optimal prediction intervals for Lipschitz predictors. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical study of regression prediction bounds, not LLM evaluation benchmarks.
Robustly Reliable Learning for Regression The authors analyze robustly reliable learning for regression under data corruptions and propose an algorithm to compute optimal prediction intervals for Lipschitz predictors. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on adversarial regression bounds rather than small models or efficient inference.
Variable Input Masked Autoencoders for Wrist-Worn Accelerometry The authors introduce a masked autoencoder that pretrains wearable sensor representations across varied recording durations and sampling rates using cross-resolution views. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Proposes masked autoencoding for sensor signals, not small LLM architectures or compression.
Variable Input Masked Autoencoders for Wrist-Worn Accelerometry The authors introduce a masked autoencoder that pretrains wearable sensor representations across varied recording durations and sampling rates using cross-resolution views. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Investigates wearable sensor encoders, not language model evaluation or benchmarks.
Variable Input Masked Autoencoders for Wrist-Worn Accelerometry The authors introduce a masked autoencoder that pretrains wearable sensor representations across varied recording durations and sampling rates using cross-resolution views. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on sensor pretraining techniques rather than deployment safety or guardrails.
Measure Neural Mappings The authors define measure neural mappings and establish quantitative error bounds for self-attention transformers acting on probability measures in unbounded spaces. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical bounds for attention over continuous measures, not model compression or efficiency.
Measure Neural Mappings The authors define measure neural mappings and establish quantitative error bounds for self-attention transformers acting on probability measures in unbounded spaces. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical operator learning without addressing AI safety, robustness, or deployment guardrails.
Measure Neural Mappings The authors define measure neural mappings and establish quantitative error bounds for self-attention transformers acting on probability measures in unbounded spaces. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Mathematical analysis of attention on probability measures, not LLM benchmark evaluation.
A Method for Exact Enforcement of Boundary Conditions on General Curved Triangular Domains: Algorithm and Application to Scientific Machine Learning The authors present a mathematical method to enforce Dirichlet, Neumann, and Robin boundary conditions exactly on curved triangular domains for scientific machine learning. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Addresses boundary value problems in numerical analysis, not LLM benchmarks.
A Method for Exact Enforcement of Boundary Conditions on General Curved Triangular Domains: Algorithm and Application to Scientific Machine Learning The authors present a mathematical method to enforce Dirichlet, Neumann, and Robin boundary conditions exactly on curved triangular domains for scientific machine learning. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on scientific machine learning on curved domains, not small models or efficient inference.
A Method for Exact Enforcement of Boundary Conditions on General Curved Triangular Domains: Algorithm and Application to Scientific Machine Learning The authors present a mathematical method to enforce Dirichlet, Neumann, and Robin boundary conditions exactly on curved triangular domains for scientific machine learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Concerns mathematical PDE solvers rather than AI deployment safety, alignment, or monitoring.
JustOPD: On-Policy Distillation from Autoregressive to Diffusion Language Models The authors introduce an on-policy distillation method to transfer knowledge from autoregressive language models into diffusion language models using autoregressive rollouts. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses model distillation mechanics rather than safety, hallucinations, or deployment guardrails.
JustOPD: On-Policy Distillation from Autoregressive to Diffusion Language Models The authors introduce an on-policy distillation method to transfer knowledge from autoregressive language models into diffusion language models using autoregressive rollouts. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Introduces a knowledge distillation technique rather than a benchmark or evaluation method.
Position: Beyond Training Time, Test-Time Coordination is Essential for Cooperative MARL The authors argue that cooperative multi-agent reinforcement learning systems require test-time coordination mechanisms rather than relying solely on training-time centralized information. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses execution frameworks in multi-agent reinforcement learning, not small or efficient models.
Position: Beyond Training Time, Test-Time Coordination is Essential for Cooperative MARL The authors argue that cooperative multi-agent reinforcement learning systems require test-time coordination mechanisms rather than relying solely on training-time centralized information. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on MARL coordination strategies rather than AI safety, red-teaming, or model alignment.
Position: Beyond Training Time, Test-Time Coordination is Essential for Cooperative MARL The authors argue that cooperative multi-agent reinforcement learning systems require test-time coordination mechanisms rather than relying solely on training-time centralized information. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes a coordination paradigm in reinforcement learning, not LLM evaluation methodologies.

9,241–9,300 of 17,313← Previous Next →