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
Optimal Transport PAC-Bayes Bounds The paper derives PAC-Bayesian generalization bounds that substitute standard relative entropy with general optimal transport metrics. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Statistical learning theory bounds, not small or efficient models.
Optimal Transport PAC-Bayes Bounds The paper derives PAC-Bayesian generalization bounds that substitute standard relative entropy with general optimal transport metrics. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical statistical learning bounds, not LLM evaluation benchmarks.
The Language of Looking: Motion Tokens Reveal Domain-Dependent Local Structure in Gaze Sequences The paper analyzes human gaze sequences across datasets using motion tokens to demonstrate domain-dependent local transition patterns. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies human gaze sequence motion tokens, not model compression.
The Language of Looking: Motion Tokens Reveal Domain-Dependent Local Structure in Gaze Sequences The paper analyzes human gaze sequences across datasets using motion tokens to demonstrate domain-dependent local transition patterns. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Gaze behavior analysis, unrelated to AI safety or guardrails.
The Language of Looking: Motion Tokens Reveal Domain-Dependent Local Structure in Gaze Sequences The paper analyzes human gaze sequences across datasets using motion tokens to demonstrate domain-dependent local transition patterns. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies gaze sequences, not language model evaluation or benchmarks.
$\texttt{NFDiag}$: High-dimensional MCMC Convergence Diagnosis with Normalizing Flows The paper proposes using normalizing flows to evaluate high-dimensional Markov Chain Monte Carlo convergence from multivariate joint posteriors. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Normalizing flows for MCMC, not small or efficient models.
$\texttt{NFDiag}$: High-dimensional MCMC Convergence Diagnosis with Normalizing Flows The paper proposes using normalizing flows to evaluate high-dimensional Markov Chain Monte Carlo convergence from multivariate joint posteriors. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on MCMC convergence diagnostics, not LLM benchmarking.
$\texttt{NFDiag}$: High-dimensional MCMC Convergence Diagnosis with Normalizing Flows The paper proposes using normalizing flows to evaluate high-dimensional Markov Chain Monte Carlo convergence from multivariate joint posteriors. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical Bayesian inference diagnostics, not deployment safety or robustness.
LLM-HB: Language-Aware LLM-Guided Hotword Biasing for Code-Switching ASR The paper presents a language-aware framework using language models and mixture-of-experts adaptors to improve speech recognition during code-switching. Peer-reviewed	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on speech recognition biasing rather than LLM evaluation benchmarks.
LLM-HB: Language-Aware LLM-Guided Hotword Biasing for Code-Switching ASR The paper presents a language-aware framework using language models and mixture-of-experts adaptors to improve speech recognition during code-switching. Peer-reviewed	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with speech recognition accuracy rather than deployment safety or guardrails.
LLM-HB: Language-Aware LLM-Guided Hotword Biasing for Code-Switching ASR The paper presents a language-aware framework using language models and mixture-of-experts adaptors to improve speech recognition during code-switching. Peer-reviewed	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses code-switching ASR biasing, not small or efficient models.
Characterizing and Mitigating Distributional Bias in Physics-Informed Generative Models The paper introduces a data-anchored penalty to prevent distributional bias and leakage in physics-informed diffusion models. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses residual regularized diffusion models, not model efficiency or compression.
Characterizing and Mitigating Distributional Bias in Physics-Informed Generative Models The paper introduces a data-anchored penalty to prevent distributional bias and leakage in physics-informed diffusion models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on physics constraints in generative models rather than deployment safety.
Characterizing and Mitigating Distributional Bias in Physics-Informed Generative Models The paper introduces a data-anchored penalty to prevent distributional bias and leakage in physics-informed diffusion models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies physics-informed generative models, not LLM benchmarks or evaluation.
OctMole: Multiscale 3D Molecular Generation with Hierarchical Spatial Flows The paper introduces a multiscale octree flow framework for generating three-dimensional molecular structures and atom placements. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Not about AI safety, hallucination, guardrails, or model alignment.
OctMole: Multiscale 3D Molecular Generation with Hierarchical Spatial Flows The paper introduces a multiscale octree flow framework for generating three-dimensional molecular structures and atom placements. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Proposes a 3D molecular flow model, not model compression or SLMs.
OctMole: Multiscale 3D Molecular Generation with Hierarchical Spatial Flows The paper introduces a multiscale octree flow framework for generating three-dimensional molecular structures and atom placements. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on 3D molecular generation rather than evaluating LLMs.
HyBRA-EEG: Hyperbolic Global-Local Brain-Region Aggregation for Self-Supervised EEG Representation Learning The paper proposes a hyperbolic aggregation framework for self-supervised brain wave representation learning across heterogeneous electrode montages. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Deals with brain-region EEG models, not efficient model serving.
HyBRA-EEG: Hyperbolic Global-Local Brain-Region Aggregation for Self-Supervised EEG Representation Learning The paper proposes a hyperbolic aggregation framework for self-supervised brain wave representation learning across heterogeneous electrode montages. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on EEG representation learning, not LLM evaluation benchmarks.
HyBRA-EEG: Hyperbolic Global-Local Brain-Region Aggregation for Self-Supervised EEG Representation Learning The paper proposes a hyperbolic aggregation framework for self-supervised brain wave representation learning across heterogeneous electrode montages. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant EEG representation learning method, not AI safety or guardrails.
Prorsa: Unleashing the Potential of Prefill-Decode Disaggregation in Agentic Reinforcement Learning The paper explores prefill-decode disaggregation architectures to speed up rollout generation during agentic reinforcement learning training. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant System infrastructure for RL rollouts, not deployment safety or robustness.
Prorsa: Unleashing the Potential of Prefill-Decode Disaggregation in Agentic Reinforcement Learning The paper explores prefill-decode disaggregation architectures to speed up rollout generation during agentic reinforcement learning training. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on serving infrastructure for RL rollout, not LLM evaluation.
DEFlow: Deep Equilibrium Flow Maximum-Entropy Reinforcement Learning The paper proposes an implicit integration approach for flow-based reinforcement learning policies to enable stable off-policy optimization. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on reinforcement learning algorithms, not LLM evaluation.
DEFlow: Deep Equilibrium Flow Maximum-Entropy Reinforcement Learning The paper proposes an implicit integration approach for flow-based reinforcement learning policies to enable stable off-policy optimization. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Implicit flow policy optimization in RL, not model compression.
DEFlow: Deep Equilibrium Flow Maximum-Entropy Reinforcement Learning The paper proposes an implicit integration approach for flow-based reinforcement learning policies to enable stable off-policy optimization. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:18	Not relevant Covers reinforcement learning continuous control, not industrial vision inspection.
DEFlow: Deep Equilibrium Flow Maximum-Entropy Reinforcement Learning The paper proposes an implicit integration approach for flow-based reinforcement learning policies to enable stable off-policy optimization. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Robotics policy optimization technique, not AI deployment safety.
DR-APO: Decoupled-Representation Advantage-Guided Policy Optimization from Continuous Control to Embodied Navigation The paper introduces an advantage-guided policy optimization framework that decouples policy representation from reinforcement learning policy updates. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on generic policy optimization, not AI safety or guardrails.
DR-APO: Decoupled-Representation Advantage-Guided Policy Optimization from Continuous Control to Embodied Navigation The paper introduces an advantage-guided policy optimization framework that decouples policy representation from reinforcement learning policy updates. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Reinforcement learning optimization method, not model compression or serving.
DR-APO: Decoupled-Representation Advantage-Guided Policy Optimization from Continuous Control to Embodied Navigation The paper introduces an advantage-guided policy optimization framework that decouples policy representation from reinforcement learning policy updates. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:18	Not relevant Deals with continuous control reinforcement learning, not industrial computer vision.
DR-APO: Decoupled-Representation Advantage-Guided Policy Optimization from Continuous Control to Embodied Navigation The paper introduces an advantage-guided policy optimization framework that decouples policy representation from reinforcement learning policy updates. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on RL policy optimization, not LLM evaluation benchmarks.
ACED: Reference-Anchored Aggregation of Natural-Language and Executable Evidence for Mathematical Reasoning The paper combines natural language reasoning with executable program evidence using reference anchoring to improve mathematical problem solving. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies aggregation for math reasoning, not model compression or efficiency.
ACED: Reference-Anchored Aggregation of Natural-Language and Executable Evidence for Mathematical Reasoning The paper combines natural language reasoning with executable program evidence using reference anchoring to improve mathematical problem solving. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on test-time aggregation of mathematical reasoning, not LLM evaluation.
ACED: Reference-Anchored Aggregation of Natural-Language and Executable Evidence for Mathematical Reasoning The paper combines natural language reasoning with executable program evidence using reference anchoring to improve mathematical problem solving. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Ensemble aggregation heuristic for math problems, not deployment safety guardrails.
Learning Deployable Causal Action Geometry under Temporal Non-Stationarity The paper presents a causal action geometry framework to estimate stable decision-response shapes from non-stationary time series. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Causal inference under non-stationarity, not efficient model architectures.
Learning Deployable Causal Action Geometry under Temporal Non-Stationarity The paper presents a causal action geometry framework to estimate stable decision-response shapes from non-stationary time series. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Causal time-series estimation method, not AI safety or red-teaming.
Learning Deployable Causal Action Geometry under Temporal Non-Stationarity The paper presents a causal action geometry framework to estimate stable decision-response shapes from non-stationary time series. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on causal decision-making under drift, not LLM evaluation.
IND-Eval: Aligning Evaluation and Simulation for Industrial Time Series Foundation Models The paper introduces an industrial time-series forecasting benchmark to evaluate foundation models on multi-rate and multivariate data. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on time-series benchmarking and simulation, not efficient model serving.
IND-Eval: Aligning Evaluation and Simulation for Industrial Time Series Foundation Models The paper introduces an industrial time-series forecasting benchmark to evaluate foundation models on multi-rate and multivariate data. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies time series benchmarking, not AI safety or guardrails.
Counting with Attention and LayerNorm: Norm Scaling and Representation Geometry The paper studies sequence-length parameter bounds and representation geometry required for attention layers to perform counting tasks. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Theoretical representation analysis of single-layer attention, not efficient model architectures.
Counting with Attention and LayerNorm: Norm Scaling and Representation Geometry The paper studies sequence-length parameter bounds and representation geometry required for attention layers to perform counting tasks. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical analysis of counting representations, not LLM evaluation benchmarks.
Counting with Attention and LayerNorm: Norm Scaling and Representation Geometry The paper studies sequence-length parameter bounds and representation geometry required for attention layers to perform counting tasks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Mathematical analysis of layer norm scaling, not deployment safety.
Guidance Without Labels: Flow Matching from Coupled Gaussian Mixtures The paper applies optimal transport with Gaussian mixtures to improve flow matching sample efficiency without requiring training labels. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Generative flow matching method, not model quantization, distillation, or serving.
Guidance Without Labels: Flow Matching from Coupled Gaussian Mixtures The paper applies optimal transport with Gaussian mixtures to improve flow matching sample efficiency without requiring training labels. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Image generation technique, not AI safety or deployment monitoring.
Guidance Without Labels: Flow Matching from Coupled Gaussian Mixtures The paper applies optimal transport with Gaussian mixtures to improve flow matching sample efficiency without requiring training labels. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Deals with flow matching image generation, not LLM evaluation.
RFWeave: Learning Wireless Propagation Across Scenes and Transceiver Configurations The paper introduces a model that predicts wireless propagation across varying physical scenes, transceiver positions, and antenna configurations. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Wireless propagation modeling paper, not language model evaluation.
RFWeave: Learning Wireless Propagation Across Scenes and Transceiver Configurations The paper introduces a model that predicts wireless propagation across varying physical scenes, transceiver positions, and antenna configurations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Wireless signal modeling, not AI safety or deployment reliability.
RFWeave: Learning Wireless Propagation Across Scenes and Transceiver Configurations The paper introduces a model that predicts wireless propagation across varying physical scenes, transceiver positions, and antenna configurations. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Applies deep learning to wireless propagation, not small efficient models.
Compression Through Time: An Interplay of Separability, Stationarity and Capacity The paper investigates information-theoretic and geometric compression mechanisms over time within recurrent neural networks. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Information-theoretic analysis of LSTM dynamics, not model compression or deployment.
Compression Through Time: An Interplay of Separability, Stationarity and Capacity The paper investigates information-theoretic and geometric compression mechanisms over time within recurrent neural networks. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies generalization dynamics in LSTMs, not LLM evaluation benchmarks.
Compression Through Time: An Interplay of Separability, Stationarity and Capacity The paper investigates information-theoretic and geometric compression mechanisms over time within recurrent neural networks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Investigates theoretical compression in recurrent networks, not safety guardrails.
Making LoRA Identifiable: Orthogonal Alignment for Continual Learning The paper introduces an orthogonal parameter alignment approach for low-rank adaptation to mitigate catastrophic forgetting in continual learning. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses catastrophic forgetting in continual learning, not deployment safety risks.
Making LoRA Identifiable: Orthogonal Alignment for Continual Learning The paper introduces an orthogonal parameter alignment approach for low-rank adaptation to mitigate catastrophic forgetting in continual learning. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on continual parameter-efficient fine-tuning, not LLM benchmark evaluation.
Amortized Variational Inference of Latent Intensities from Replicated Point Patterns The paper develops an amortized variational inference method to estimate latent point process intensities from replicated event patterns. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies latent point processes, not language model evaluation or benchmarks.
Amortized Variational Inference of Latent Intensities from Replicated Point Patterns The paper develops an amortized variational inference method to estimate latent point process intensities from replicated event patterns. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Concerns spatial-temporal point processes, not deployment AI safety.
Amortized Variational Inference of Latent Intensities from Replicated Point Patterns The paper develops an amortized variational inference method to estimate latent point process intensities from replicated event patterns. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on Bayesian point process inference, not LLM efficiency.
Spectral Costs of Multinomial and Residual Resampling in High-Dimensional Particle Ensembles: A Free-Probability Analysis The paper analyzes the spectral properties and limits of multinomial and residual resampling in high-dimensional particle filtering. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Mathematical analysis of particle ensembles, not efficient model inference.
Spectral Costs of Multinomial and Residual Resampling in High-Dimensional Particle Ensembles: A Free-Probability Analysis The paper analyzes the spectral properties and limits of multinomial and residual resampling in high-dimensional particle filtering. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies particle filter resampling theory, not language model evaluation.
Spectral Costs of Multinomial and Residual Resampling in High-Dimensional Particle Ensembles: A Free-Probability Analysis The paper analyzes the spectral properties and limits of multinomial and residual resampling in high-dimensional particle filtering. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Mathematical theory of particle filters, not AI safety deployment.
Universal Distributed Feature Extraction Complementary to Side Information via H-Score The paper introduces a nested H-score method for distributed feature extraction that avoids redundancy with receiver-side auxiliary information. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on distributed estimation theory, not AI safety guardrails.
Universal Distributed Feature Extraction Complementary to Side Information via H-Score The paper introduces a nested H-score method for distributed feature extraction that avoids redundancy with receiver-side auxiliary information. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Deals with information theory in distributed estimation, not LLM evaluation.

5,521–5,580 of 13,754← Previous Next →