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
Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State Quantization This paper redesigns 4-bit AdamW optimizer-state quantization by performing stochastic rounding directly within preconditioner space. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies optimizer-state quantization rather than language model evaluation methods.
Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State Quantization This paper redesigns 4-bit AdamW optimizer-state quantization by performing stochastic rounding directly within preconditioner space. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with numerical optimization and quantization, not AI deployment safety or robustness.
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control This paper analyzes data sampling distributions across parallel simulations to improve exploration efficiency in reinforcement learning for robot control. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Addresses simulation sample efficiency in RL, not language model evaluation.
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control This paper analyzes data sampling distributions across parallel simulations to improve exploration efficiency in reinforcement learning for robot control. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:29	Not relevant Investigates RL exploration sampling for robotic control rather than industrial vision inspection.
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control This paper analyzes data sampling distributions across parallel simulations to improve exploration efficiency in reinforcement learning for robot control. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on RL training sample selection, not small or efficient model architectures.
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control This paper analyzes data sampling distributions across parallel simulations to improve exploration efficiency in reinforcement learning for robot control. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Targets general robot control learning rather than deployment safety constraints or guardrails.
CSF: Contextual Safety Filtering for Motion Generators This paper introduces a training-free safety filter that evaluates scene context to constrain generated humanoid motion trajectories. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Proposes a control safety filter for robot motion, not LLM benchmark evaluation.
CSF: Contextual Safety Filtering for Motion Generators This paper introduces a training-free safety filter that evaluates scene context to constrain generated humanoid motion trajectories. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:29	Not relevant Focuses on robotic motion generation safety rather than industrial computer vision inspection.
CSF: Contextual Safety Filtering for Motion Generators This paper introduces a training-free safety filter that evaluates scene context to constrain generated humanoid motion trajectories. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Does not address model compression, quantization, or efficient model serving.
Sample-Adaptive Zeroth-Order Post-Training This paper proposes a sample-adaptive zeroth-order optimization method that estimates sample importance online to accelerate large language model fine-tuning. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on optimization convergence and memory efficiency rather than AI safety or reliability.
Sample-Adaptive Zeroth-Order Post-Training This paper proposes a sample-adaptive zeroth-order optimization method that estimates sample importance online to accelerate large language model fine-tuning. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Develops a post-training optimization method without focusing on LLM evaluation.
An all-transistor physical-learning transformer can be trained to perform in-context learning This paper designs and simulates an all-transistor physical circuit using sub-threshold MOSFETs to run transformer attention and in-context learning. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Studies circuit design for physical learning, not deployment safety or reliability.
An all-transistor physical-learning transformer can be trained to perform in-context learning This paper designs and simulates an all-transistor physical circuit using sub-threshold MOSFETs to run transformer attention and in-context learning. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on hardware-level physical learning circuits rather than LLM benchmarks or evaluation.
DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models This paper proposes a reward aggregation method to stabilize and scale reinforcement learning fine-tuning for diffusion models. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Targets RL fine-tuning stability for diffusion generation, not model efficiency or compression.
DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models This paper proposes a reward aggregation method to stabilize and scale reinforcement learning fine-tuning for diffusion models. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Develops an RL algorithm for diffusion models without proposing LLM evaluation frameworks.
DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models This paper proposes a reward aggregation method to stabilize and scale reinforcement learning fine-tuning for diffusion models. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on general diffusion model RL optimization, not safety, guardrails, or alignment.
Realism VS Accuracy: Event Sequence Forecasting from a Generative Modeling Perspective This paper introduces a benchmark and a diffusion aggregation algorithm to evaluate and balance accuracy and realism in event sequence forecasting. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses forecast realism versus accuracy, not AI deployment safety, guardrails, or alignment.
Realism VS Accuracy: Event Sequence Forecasting from a Generative Modeling Perspective This paper introduces a benchmark and a diffusion aggregation algorithm to evaluate and balance accuracy and realism in event sequence forecasting. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on event sequence generative modeling, not model compression or efficient serving.
Realism VS Accuracy: Event Sequence Forecasting from a Generative Modeling Perspective This paper introduces a benchmark and a diffusion aggregation algorithm to evaluate and balance accuracy and realism in event sequence forecasting. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Evaluates event sequence forecasting models, not large language models or LLM evaluation.
STAPLE: Static Sparse Upcycling of Pretrained Checkpoints via Embedding Modules This paper presents a method for upcycling pretrained language models into sparse architectures using token-indexed embeddings in residual pathways. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on sparse pretraining architectures, not AI safety, robustness, or reliability.
STAPLE: Static Sparse Upcycling of Pretrained Checkpoints via Embedding Modules This paper presents a method for upcycling pretrained language models into sparse architectures using token-indexed embeddings in residual pathways. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Evaluates on standard benchmarks but does not introduce or study LLM evaluation methods.
SynAct: Synergistic Active Feedback Learning for LLM Agents This paper proposes an active learning framework that selectively requests expert feedback at high-value decision steps during LLM agent workflows. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on active learning algorithms for agent workflows, not LLM benchmark evaluation.
SynAct: Synergistic Active Feedback Learning for LLM Agents This paper proposes an active learning framework that selectively requests expert feedback at high-value decision steps during LLM agent workflows. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on active feedback mechanisms for agents, not model compression or efficiency.
Data-driven Optimal Sensor Placement for Aerosol Monitoring in the UK This paper applies a convolutional Gaussian neural process and genetic optimization to select optimal aerosol monitoring sensor locations across the UK. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on environmental sensor optimization, not small or efficient language models.
Data-driven Optimal Sensor Placement for Aerosol Monitoring in the UK This paper applies a convolutional Gaussian neural process and genetic optimization to select optimal aerosol monitoring sensor locations across the UK. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Covers sensor placement for atmospheric monitoring, not LLM evaluation.
Data-driven Optimal Sensor Placement for Aerosol Monitoring in the UK This paper applies a convolutional Gaussian neural process and genetic optimization to select optimal aerosol monitoring sensor locations across the UK. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with environmental sensor network optimization, not AI safety or guardrails.
Amortised Inference through One-Step Implicit Sampling This paper introduces a one-step implicit sampler using short-time MCMC dynamics to enable efficient amortized inference for Bayesian posteriors. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on high-dimensional Bayesian sampling algorithms, not small or efficient LLMs.
Amortised Inference through One-Step Implicit Sampling This paper introduces a one-step implicit sampler using short-time MCMC dynamics to enable efficient amortized inference for Bayesian posteriors. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Addresses Bayesian posterior sampling techniques, not LLM evaluation.
Amortised Inference through One-Step Implicit Sampling This paper introduces a one-step implicit sampler using short-time MCMC dynamics to enable efficient amortized inference for Bayesian posteriors. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on implicit posterior sampling, not AI deployment safety or reliability.
Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy This paper uses a ConvNeXt architecture to detect low-voltage EEG intervals as biomarkers for neonatal hypoxic-ischaemic encephalopathy. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Applies ConvNeXt to neonatal EEG data, not LLM evaluation or benchmarks.
Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy This paper uses a ConvNeXt architecture to detect low-voltage EEG intervals as biomarkers for neonatal hypoxic-ischaemic encephalopathy. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on clinical EEG classification, not small or efficient language models.
Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy This paper uses a ConvNeXt architecture to detect low-voltage EEG intervals as biomarkers for neonatal hypoxic-ischaemic encephalopathy. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with medical signal detection, not AI safety or deployment guardrails.
Critical Mass: Structure-Disjoint Splits Drift With the Libraries They Are Built On This paper optimizes the Maximum Common Edge Subgraph distance calculation to evaluate structure-disjoint data splits on large molecular libraries. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies dataset split leakage in cheminformatics, not LLM evaluation or benchmarks.
Critical Mass: Structure-Disjoint Splits Drift With the Libraries They Are Built On This paper optimizes the Maximum Common Edge Subgraph distance calculation to evaluate structure-disjoint data splits on large molecular libraries. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Addresses molecular graph partitioning, not small language models or model efficiency.
Critical Mass: Structure-Disjoint Splits Drift With the Libraries They Are Built On This paper optimizes the Maximum Common Edge Subgraph distance calculation to evaluate structure-disjoint data splits on large molecular libraries. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on cheminformatics splitting algorithms, not AI deployment safety.
Toward Maximizing Algebraic Connectivity via Density-Dependent Backbone Construction and Reinforcement Learning Completion Combines structural graph backbone construction with reinforcement learning completion to design network topologies that maximize algebraic connectivity. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on graph topology optimization, not small or efficient language models.
Toward Maximizing Algebraic Connectivity via Density-Dependent Backbone Construction and Reinforcement Learning Completion Combines structural graph backbone construction with reinforcement learning completion to design network topologies that maximize algebraic connectivity. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Applies RL to graph connectivity problems, not LLM evaluation.
Toward Maximizing Algebraic Connectivity via Density-Dependent Backbone Construction and Reinforcement Learning Completion Combines structural graph backbone construction with reinforcement learning completion to design network topologies that maximize algebraic connectivity. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on graph theory and network topology design, not AI safety or reliability.
Primal-Dual Flow Matching for Sample-Wise Constrained Generation Introduces primal-dual flow matching to enforce sample-wise constraints in generative modeling using binary membership feedback. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses mathematical constraint satisfaction in flow matching, not LLM deployment safety.
Primal-Dual Flow Matching for Sample-Wise Constrained Generation Introduces primal-dual flow matching to enforce sample-wise constraints in generative modeling using binary membership feedback. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on flow-matching generative models, not LLM evaluation or benchmarks.
Primal-Dual Flow Matching for Sample-Wise Constrained Generation Introduces primal-dual flow matching to enforce sample-wise constraints in generative modeling using binary membership feedback. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Deals with constrained continuous generative flow models, not small or efficient LLMs.
Product Monte Carlo: Expectation Under a Product of Densities Proposes Product Monte Carlo to compute expectations under products of densities using draws from individual densities via regression. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses general Bayesian statistical estimation, not AI deployment safety or guardrails.
Product Monte Carlo: Expectation Under a Product of Densities Proposes Product Monte Carlo to compute expectations under products of densities using draws from individual densities via regression. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on Bayesian Monte Carlo expectation estimation, not LLM evaluation.
Product Monte Carlo: Expectation Under a Product of Densities Proposes Product Monte Carlo to compute expectations under products of densities using draws from individual densities via regression. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Deals with statistical density aggregation, not small or efficient language models.
A Distributional Framework for Behavioral Characterization and Risk Management of Large Language Models Presents a Monte Carlo simulation framework to quantify distributional response shifts and model risks across language model lifecycles. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on behavioral risk characterization, not model compression or efficiency.
Spatial Context or Memory? Sizing the Inputs of Neural Surrogates for Partially Observed Fields Analyzes the optimal balance between spatial sensor window size and temporal history when forecasting partially observed physical fields. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on neural surrogates in fluid mechanics, not AI safety or deployment guardrails.
Spatial Context or Memory? Sizing the Inputs of Neural Surrogates for Partially Observed Fields Analyzes the optimal balance between spatial sensor window size and temporal history when forecasting partially observed physical fields. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on physics surrogates, not small language models or efficient LLM inference.
Spatial Context or Memory? Sizing the Inputs of Neural Surrogates for Partially Observed Fields Analyzes the optimal balance between spatial sensor window size and temporal history when forecasting partially observed physical fields. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies physical field forecasting surrogates, not LLM benchmarks or evaluation methods.
A Categorical Atlas of Parameter-Efficient Fine-Tuning: Cones, Orbits, and Gauges Develops a category-theoretic framework to analyze and compare parameter-efficient fine-tuning methods as smooth maps in weight space. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Theoretical analysis of fine-tuning parameter spaces, not safety or deployment reliability.
A Categorical Atlas of Parameter-Efficient Fine-Tuning: Cones, Orbits, and Gauges Develops a category-theoretic framework to analyze and compare parameter-efficient fine-tuning methods as smooth maps in weight space. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Theoretical category theory paper on PEFT spaces, not LLM evaluation or benchmarking.
Minimax Rates for Estimating Powered Sliced Wasserstein Functionals Establishes minimax estimation rates and examines plug-in estimators for powered sliced Wasserstein functionals under moment bounds. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Pure mathematical statistics paper on Wasserstein distances, not LLM evaluation.
Minimax Rates for Estimating Powered Sliced Wasserstein Functionals Establishes minimax estimation rates and examines plug-in estimators for powered sliced Wasserstein functionals under moment bounds. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with theoretical statistical estimation rates, not AI safety or deployment risks.
Minimax Rates for Estimating Powered Sliced Wasserstein Functionals Establishes minimax estimation rates and examines plug-in estimators for powered sliced Wasserstein functionals under moment bounds. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Covers statistical convergence rates rather than small or efficient language models.
Neural expectation-maximization and meta-models Interprets Transformer networks as learned approximations to loopy belief propagation and expectation-maximization in dynamic Markov random fields. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Theoretical analysis of Transformer inference mechanisms, not model compression or efficient serving.
Neural expectation-maximization and meta-models Interprets Transformer networks as learned approximations to loopy belief propagation and expectation-maximization in dynamic Markov random fields. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on probabilistic interpretations of self-attention, not deployment safety or robustness.
Neural expectation-maximization and meta-models Interprets Transformer networks as learned approximations to loopy belief propagation and expectation-maximization in dynamic Markov random fields. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents a probabilistic theoretical formulation of Transformers, not LLM evaluation or benchmarking.
Static Gradient Attribution Underperforms a Density-Matched Random Mask on Loss-Based Forgetting Within LoRA B-Matrix Reports that initialization-time gradient masks underperform random masks on catastrophic forgetting when confined to the same LoRA parameter pool. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Investigates continual learning forgetting in LoRA, not LLM benchmark design or evaluation methodology.
Static Gradient Attribution Underperforms a Density-Matched Random Mask on Loss-Based Forgetting Within LoRA B-Matrix Reports that initialization-time gradient masks underperform random masks on catastrophic forgetting when confined to the same LoRA parameter pool. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Studies parameter selection for continual learning rather than AI safety, red-teaming, or guardrails.
NV-Tesseract-AD: Curriculum-Enhanced Conditional Diffusion for Multivariate Time-Series Anomaly Detection Introduces a dual-curriculum conditional diffusion framework for multivariate time-series anomaly detection across varying masking and noise levels. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on time-series diffusion models rather than efficient language model serving.
NV-Tesseract-AD: Curriculum-Enhanced Conditional Diffusion for Multivariate Time-Series Anomaly Detection Introduces a dual-curriculum conditional diffusion framework for multivariate time-series anomaly detection across varying masking and noise levels. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Evaluates time-series anomaly detection algorithms, not LLMs or LLM benchmarks.

11,341–11,400 of 13,754← Previous Next →