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
Toward a Multimodal Foundation Model for Plasma State Reconstruction Reconstructs magnetic reconnection plasma states from sparse multimodal observations using a residual 3D U-Net architecture. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies plasma physics reconstruction, not deployment safety risks.
On the Theoretical Limits of Learning Unconstrained Interatomic Potentials Demonstrates theoretical sampling failures and kinetic traps when training unconstrained machine-learned interatomic potentials from energy and force data. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers physical molecular dynamics, not AI safety and reliability.
On the Theoretical Limits of Learning Unconstrained Interatomic Potentials Demonstrates theoretical sampling failures and kinetic traps when training unconstrained machine-learned interatomic potentials from energy and force data. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical analysis of molecular potentials, not LLM benchmarks.
On the Theoretical Limits of Learning Unconstrained Interatomic Potentials Demonstrates theoretical sampling failures and kinetic traps when training unconstrained machine-learned interatomic potentials from energy and force data. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies interatomic potentials, not model compression or efficiency.
Swarm Telepathy: Prefill-Free KV-Cache Transfer Across Families Enables prefill-free multi-agent communication across different language model families using cross-model key-value cache transfer. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on inference speedup, not safety, guardrails, or hallucination.
Swarm Telepathy: Prefill-Free KV-Cache Transfer Across Families Enables prefill-free multi-agent communication across different language model families using cross-model key-value cache transfer. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes KV-cache transfer, not benchmark evaluation methodologies.
Anchors Enhance Controllability and Reasoning of Diffusion Language Models Uses reinforcement learning to place token anchors that guide text generation in frozen diffusion language models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on generation controllability in DLMs, not deployment safety risks.
Anchors Enhance Controllability and Reasoning of Diffusion Language Models Uses reinforcement learning to place token anchors that guide text generation in frozen diffusion language models. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies diffusion language model controllability, not model compression.
Anchors Enhance Controllability and Reasoning of Diffusion Language Models Uses reinforcement learning to place token anchors that guide text generation in frozen diffusion language models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes anchor-guided generation, not LLM benchmark evaluation.
Random Feature Pruning for Multi-Class Classification Proposes a sequential selection algorithm to prune redundant random features for multiclass classification with theoretical guarantees. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies feature pruning algorithms, not AI safety or alignment.
Random Feature Pruning for Multi-Class Classification Proposes a sequential selection algorithm to prune redundant random features for multiclass classification with theoretical guarantees. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on classical random feature pruning, not small language models.
Random Feature Pruning for Multi-Class Classification Proposes a sequential selection algorithm to prune redundant random features for multiclass classification with theoretical guarantees. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies random kitchen sinks pruning, not LLM benchmarks.
RELAGRAM: Recursive Latent Graph Model for the Capacitated Vehicle Routing Problem Introduces a recurrent latent graph transformer that solves capacitated vehicle routing problems by updating edge representations iteratively. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Applies latent recursion to CVRP, not LLM evaluation.
RELAGRAM: Recursive Latent Graph Model for the Capacitated Vehicle Routing Problem Introduces a recurrent latent graph transformer that solves capacitated vehicle routing problems by updating edge representations iteratively. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on combinatorial optimization, not LLM compression or inference efficiency.
RELAGRAM: Recursive Latent Graph Model for the Capacitated Vehicle Routing Problem Introduces a recurrent latent graph transformer that solves capacitated vehicle routing problems by updating edge representations iteratively. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers vehicle routing optimization, not AI safety in deployment.
Energy-Guided Learning for Heat Equation with Noisy Source: Optimal Estimation and Generalization Bounds Presents an energy-regularized regression method to learn heat equation solutions from noisy physical data and proves estimation error bounds. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical PDE learning work, not AI safety or robustness.
Energy-Guided Learning for Heat Equation with Noisy Source: Optimal Estimation and Generalization Bounds Presents an energy-regularized regression method to learn heat equation solutions from noisy physical data and proves estimation error bounds. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Studies PDE mathematical learning bounds, not model efficiency.
Energy-Guided Learning for Heat Equation with Noisy Source: Optimal Estimation and Generalization Bounds Presents an energy-regularized regression method to learn heat equation solutions from noisy physical data and proves estimation error bounds. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on solving heat equations, not LLM benchmarks.
Data-Topology Compatibility Enables Personalization in Decentralized Learning Analyzes how data and network structure alignment enables personalized model groups in decentralized learning without explicit personalization algorithms. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical study of decentralized learning, not deployment safety risks.
Data-Topology Compatibility Enables Personalization in Decentralized Learning Analyzes how data and network structure alignment enables personalized model groups in decentralized learning without explicit personalization algorithms. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical decentralized learning, not efficient model architectures.
Data-Topology Compatibility Enables Personalization in Decentralized Learning Analyzes how data and network structure alignment enables personalized model groups in decentralized learning without explicit personalization algorithms. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Analyzes decentralized learning theory, not LLM benchmark evaluation.
When Dynamic Scaling Fails: A Scale-State Feedback Governs Recurrent-State Quantization in Linear-Attention and SSM LLMs This paper investigates error compounding and scale-state feedback mechanisms during recurrent-state quantization in linear-attention and state-space language models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies recurrent state quantization failure, not LLM evaluation benchmarks.
When Dynamic Scaling Fails: A Scale-State Feedback Governs Recurrent-State Quantization in Linear-Attention and SSM LLMs This paper investigates error compounding and scale-state feedback mechanisms during recurrent-state quantization in linear-attention and state-space language models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on quantization dynamics, not safety guardrails or jailbreaks.
Sensor-Aware Joint Data Fusion and State Estimation under Sparse Multi-rate Distribution Grid Sensing This paper proposes a graph differential equation and transformer architecture to perform joint data fusion and state estimation in power distribution grids. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses power distribution sensing, not model efficiency techniques.
Sensor-Aware Joint Data Fusion and State Estimation under Sparse Multi-rate Distribution Grid Sensing This paper proposes a graph differential equation and transformer architecture to perform joint data fusion and state estimation in power distribution grids. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Studies electrical grid sensing, not AI safety and reliability.
Sensor-Aware Joint Data Fusion and State Estimation under Sparse Multi-rate Distribution Grid Sensing This paper proposes a graph differential equation and transformer architecture to perform joint data fusion and state estimation in power distribution grids. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on power grid state estimation, not LLM benchmarks.
Beyond Adjacent Layers: Graph-Guided Layer Fusion for Compressing Large Language Models This paper introduces a graph-guided layer fusion method that identifies and merges functionally redundant non-adjacent layers to compress language models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on depth compression, not benchmark evaluation design.
Beyond Adjacent Layers: Graph-Guided Layer Fusion for Compressing Large Language Models This paper introduces a graph-guided layer fusion method that identifies and merges functionally redundant non-adjacent layers to compress language models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Evaluates model pruning and fusion, not AI safety risks.
LexiCan: Lexicographic Canonicalization of Point Clouds for Machine Learning Applications This paper presents an invariant lexicographic canonicalization method to resolve rotation and permutation ambiguities in 3D point cloud representations. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Proposes point cloud canonicalization, not small or efficient models.
LexiCan: Lexicographic Canonicalization of Point Clouds for Machine Learning Applications This paper presents an invariant lexicographic canonicalization method to resolve rotation and permutation ambiguities in 3D point cloud representations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses 3D point clouds, not AI deployment safety.
LexiCan: Lexicographic Canonicalization of Point Clouds for Machine Learning Applications This paper presents an invariant lexicographic canonicalization method to resolve rotation and permutation ambiguities in 3D point cloud representations. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies 3D point cloud canonicalization, not LLM evaluation.
Physics Foundation Models for Efficient Control of Fluid Flows This paper attaches reinforcement learning control heads to a pretrained physics foundation model for closed-loop fluid-flow control. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Studies fluid flow control, not LLM evaluation benchmarks.
Physics Foundation Models for Efficient Control of Fluid Flows This paper attaches reinforcement learning control heads to a pretrained physics foundation model for closed-loop fluid-flow control. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Explores fluid flow control, not model compression or efficiency.
Physics Foundation Models for Efficient Control of Fluid Flows This paper attaches reinforcement learning control heads to a pretrained physics foundation model for closed-loop fluid-flow control. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on physics control systems, not deployment safety guardrails.
Game-Theoretic Reinforcement Learning based Topology-Agnostic Defense for Virtual Power Plants This paper proposes a game-theoretic reinforcement learning framework to defend virtual power plant resources without requiring grid network topology knowledge. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Examines electrical grid cybersecurity, not LLM safety or deployment risks.
Game-Theoretic Reinforcement Learning based Topology-Agnostic Defense for Virtual Power Plants This paper proposes a game-theoretic reinforcement learning framework to defend virtual power plant resources without requiring grid network topology knowledge. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Addresses grid cyber-physical security, not model compression or efficiency.
Game-Theoretic Reinforcement Learning based Topology-Agnostic Defense for Virtual Power Plants This paper proposes a game-theoretic reinforcement learning framework to defend virtual power plant resources without requiring grid network topology knowledge. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on power grid defense, not LLM benchmarks.
DeltaNet-PC: Test-Time Memory Without Backpropagation Through Time This paper formulates sequential test-time memory updates as latent variables within a predictive coding framework without backpropagation through time. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes a predictive coding architecture, not an evaluation benchmark.
DeltaNet-PC: Test-Time Memory Without Backpropagation Through Time This paper formulates sequential test-time memory updates as latent variables within a predictive coding framework without backpropagation through time. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on sequence memory architectures, not deployment safety.
Non-Euclidean Gradient Descent Has Many Edges of Stability This paper studies stability thresholds and convergence boundaries of non-Euclidean gradient descent on quadratic objectives. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical optimization paper, not an LLM benchmark.
Non-Euclidean Gradient Descent Has Many Edges of Stability This paper studies stability thresholds and convergence boundaries of non-Euclidean gradient descent on quadratic objectives. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Gradient descent convergence analysis, not efficient model serving.
Non-Euclidean Gradient Descent Has Many Edges of Stability This paper studies stability thresholds and convergence boundaries of non-Euclidean gradient descent on quadratic objectives. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical analysis of optimization dynamics, not AI safety.
Zeroth-Order Fine-Tuning of LLMs in Learnable Low-Rank Subspaces This paper proposes a zeroth-order fine-tuning method for language models that optimizes within task-adaptive learnable low-rank subspaces. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Zeroth-order optimization method, not an LLM benchmark.
Zeroth-Order Fine-Tuning of LLMs in Learnable Low-Rank Subspaces This paper proposes a zeroth-order fine-tuning method for language models that optimizes within task-adaptive learnable low-rank subspaces. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on training optimization, not deployment safety or robustness.
Explicit Bounds on Hoffman Constants for Sparse Polyhedral Regularizers This paper derives theoretical bounds on Hoffman constants to analyze the conditioning and convergence of sparse regularized optimization problems. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Theoretical regularizer bounds, not efficient model deployment.
Explicit Bounds on Hoffman Constants for Sparse Polyhedral Regularizers This paper derives theoretical bounds on Hoffman constants to analyze the conditioning and convergence of sparse regularized optimization problems. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Pure mathematical optimization theory, not an LLM benchmark.
Explicit Bounds on Hoffman Constants for Sparse Polyhedral Regularizers This paper derives theoretical bounds on Hoffman constants to analyze the conditioning and convergence of sparse regularized optimization problems. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Mathematical analysis of optimization, not AI deployment safety.
ObsSpec: Accelerating Agentic Reinforcement Learning with Observation Speculation This paper introduces an observation speculation framework using small world models to accelerate reinforcement learning rollouts for tool-using agents. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Inference acceleration method, not an LLM benchmark.
ObsSpec: Accelerating Agentic Reinforcement Learning with Observation Speculation This paper introduces an observation speculation framework using small world models to accelerate reinforcement learning rollouts for tool-using agents. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on training latency reduction, not AI safety.
Correlation-Aware Reconstruction for Multivariate Time Series Anomaly Detection This paper introduces a dual-latent variational autoencoder that separates temporal dynamics from cross-sensor correlations for time-series anomaly detection. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Time series anomaly detection, not LLM evaluation benchmarks.
Correlation-Aware Reconstruction for Multivariate Time Series Anomaly Detection This paper introduces a dual-latent variational autoencoder that separates temporal dynamics from cross-sensor correlations for time-series anomaly detection. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Multivariate sensor anomaly detection, not AI deployment safety.
Correlation-Aware Reconstruction for Multivariate Time Series Anomaly Detection This paper introduces a dual-latent variational autoencoder that separates temporal dynamics from cross-sensor correlations for time-series anomaly detection. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Time series reconstruction VAE, not efficient language models.
Imperfect Verifiers, Exact Targets: Inference-Time Scaling for Continuous Generation This paper introduces an inference-time scaling method using paired denoising and renoising to sample from target distributions despite imperfect verifiers. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Continuous generative sampling theory, not efficient model compression.
Imperfect Verifiers, Exact Targets: Inference-Time Scaling for Continuous Generation This paper introduces an inference-time scaling method using paired denoising and renoising to sample from target distributions despite imperfect verifiers. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Sampling theory for continuous models, not deployment safety.
Imperfect Verifiers, Exact Targets: Inference-Time Scaling for Continuous Generation This paper introduces an inference-time scaling method using paired denoising and renoising to sample from target distributions despite imperfect verifiers. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Inference-time sampling algorithm, not an LLM benchmark.
What a Data Release Must Preserve for Off-Policy Evaluation This paper derives theoretical conditions on what decision-time history data releases must retain to ensure valid off-policy evaluation. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Statistical release theory, not small efficient language models.
What a Data Release Must Preserve for Off-Policy Evaluation This paper derives theoretical conditions on what decision-time history data releases must retain to ensure valid off-policy evaluation. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical off-policy evaluation statistics, not LLM benchmark study.
What a Data Release Must Preserve for Off-Policy Evaluation This paper derives theoretical conditions on what decision-time history data releases must retain to ensure valid off-policy evaluation. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Statistical policy theory, not AI safety or deployment guardrails.
FragFlex: Fragment-Based Flexible Molecular Generation for Exploring Vast Chemical Space This paper presents a discrete diffusion model that operates at the fragment level to generate variable-sized molecular graphs. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on drug discovery diffusion, not efficient LLM serving.
FragFlex: Fragment-Based Flexible Molecular Generation for Exploring Vast Chemical Space This paper presents a discrete diffusion model that operates at the fragment level to generate variable-sized molecular graphs. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Chemistry generative model, not AI deployment safety.

5,161–5,220 of 13,754← Previous Next →