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
Harness Policy Distillation: Aligning Agent Policies with Runtime Semantics The paper introduces a distillation method that trains agent policies on explicit runtime control semantics recovered from replayed execution traces. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on harness policy distillation rather than model compression techniques.
FUSE: Fairness-aware Update Screening against Byzantine Exploitation in $q$-Fair Federated Learning The paper studies Byzantine robustness in fair federated learning and proposes an update screening method using client trajectory consistency. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Addresses federated learning robustness, not model compression or inference.
FUSE: Fairness-aware Update Screening against Byzantine Exploitation in $q$-Fair Federated Learning The paper studies Byzantine robustness in fair federated learning and proposes an update screening method using client trajectory consistency. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Focuses on federated learning optimization, not autonomous AI agents.
FUSE: Fairness-aware Update Screening against Byzantine Exploitation in $q$-Fair Federated Learning The paper studies Byzantine robustness in fair federated learning and proposes an update screening method using client trajectory consistency. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes defense screening against Byzantine updates, not LLM benchmarking.
CSWBench: Evaluating Context-Sensitive Code-Switching in Creative Writing The paper introduces a benchmark and evaluation framework assessing context-sensitive code-switching by language models in creative writing tasks. Preprint	LLMs for code via arxiv · 15 Oct, 15:06	Not relevant Addresses linguistic code-switching in creative writing, not software code.
CSWBench: Evaluating Context-Sensitive Code-Switching in Creative Writing The paper introduces a benchmark and evaluation framework assessing context-sensitive code-switching by language models in creative writing tasks. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Assesses multilingual text generation, not autonomous tool-using agents.
CSWBench: Evaluating Context-Sensitive Code-Switching in Creative Writing The paper introduces a benchmark and evaluation framework assessing context-sensitive code-switching by language models in creative writing tasks. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 15:06	Not relevant Evaluates multilingual creative writing, not retrieval-augmented generation.
CSWBench: Evaluating Context-Sensitive Code-Switching in Creative Writing The paper introduces a benchmark and evaluation framework assessing context-sensitive code-switching by language models in creative writing tasks. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Examines multilingual text generation without any visual or audio modalities.
CSWBench: Evaluating Context-Sensitive Code-Switching in Creative Writing The paper introduces a benchmark and evaluation framework assessing context-sensitive code-switching by language models in creative writing tasks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on creative writing evaluation, not deployment safety risks.
CSWBench: Evaluating Context-Sensitive Code-Switching in Creative Writing The paper introduces a benchmark and evaluation framework assessing context-sensitive code-switching by language models in creative writing tasks. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Presents an evaluation benchmark, not model efficiency techniques.
CASI: Credit Assignment via Set Intervention for Agentic Retrieval The paper presents an intervention method using Shapley values to assign credit to individual actions within multi-action agentic retrieval sets. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes an attribution algorithm, not benchmark evaluation methodologies.
CASI: Credit Assignment via Set Intervention for Agentic Retrieval The paper presents an intervention method using Shapley values to assign credit to individual actions within multi-action agentic retrieval sets. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on retrieval policy optimization, not model compression or efficiency.
CASI: Credit Assignment via Set Intervention for Agentic Retrieval The paper presents an intervention method using Shapley values to assign credit to individual actions within multi-action agentic retrieval sets. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on retrieval credit assignment, not safety or guardrails.
Reflect After Failure: Brain-Inspired Self-Evolving Agents for Embodied Spatial Intelligence The paper introduces a brain-inspired reflection framework that diagnoses failure points to guide self-evolving agents in embodied spatial tasks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on embodied spatial intelligence, not deployment safety guardrails.
Reflect After Failure: Brain-Inspired Self-Evolving Agents for Embodied Spatial Intelligence The paper introduces a brain-inspired reflection framework that diagnoses failure points to guide self-evolving agents in embodied spatial tasks. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Presents an agent architecture rather than an evaluation benchmark.
Beyond Static Cues: Unified Dynamic Detection of MLLM Backdoors via Dual-Flow Trajectory Analysis The paper detects backdoor triggers in multimodal language models by tracking output evolution trajectories across internal hidden layers. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes an attack detection framework, not a general benchmark.
Beyond Static Cues: Unified Dynamic Detection of MLLM Backdoors via Dual-Flow Trajectory Analysis The paper detects backdoor triggers in multimodal language models by tracking output evolution trajectories across internal hidden layers. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Studies backdoor defenses in MLLMs, not autonomous planning agents.
Beyond Static Cues: Unified Dynamic Detection of MLLM Backdoors via Dual-Flow Trajectory Analysis The paper detects backdoor triggers in multimodal language models by tracking output evolution trajectories across internal hidden layers. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Focuses on multimodal safety attacks, not industrial manufacturing inspection.
Argument Graph Uncertainty: Quantifying Uncertainty from the Logical Structure of Reasoning Chains The paper estimates reasoning confidence by segmenting chain-of-thought traces into argument graphs evaluated through natural language inference. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Focuses on QA reasoning chain uncertainty estimation, not agents.
Argument Graph Uncertainty: Quantifying Uncertainty from the Logical Structure of Reasoning Chains The paper estimates reasoning confidence by segmenting chain-of-thought traces into argument graphs evaluated through natural language inference. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Introduces an uncertainty estimation method rather than benchmark evaluation.
TreeFlex: Agents Navigating Their Own Search Trees The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 15:06	Not relevant Addresses agent search tree navigation, not retrieval-augmented generation systems.
TreeFlex: Agents Navigating Their Own Search Trees The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on agent search context management, not model compression.
TreeFlex: Agents Navigating Their Own Search Trees The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on agent task planning, not deployment safety guardrails.
TreeFlex: Agents Navigating Their Own Search Trees The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Deals with reasoning search trees, not multimodal vision-language tasks.
TreeFlex: Agents Navigating Their Own Search Trees The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles. Preprint	LLMs for code via arxiv · 15 Oct, 15:06	Not relevant Focuses on general tree-search agent policy, not code-specific assistants.
TreeFlex: Agents Navigating Their Own Search Trees The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes an agent framework, not benchmark evaluation methodologies.
Read, Parse, Describe: Unified Document Parsing with Visual Element Description Generation The paper presents a benchmark dataset evaluating unified page-level document parsing alongside alternative text descriptions for visual elements. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Focuses on document parsing and captioning, not autonomous agents.
Read, Parse, Describe: Unified Document Parsing with Visual Element Description Generation The paper presents a benchmark dataset evaluating unified page-level document parsing alongside alternative text descriptions for visual elements. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Addresses general document understanding, not industrial manufacturing defect inspection.
Read, Parse, Describe: Unified Document Parsing with Visual Element Description Generation The paper presents a benchmark dataset evaluating unified page-level document parsing alongside alternative text descriptions for visual elements. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on multimodal document layout parsing, not safety guardrails.
Read, Parse, Describe: Unified Document Parsing with Visual Element Description Generation The paper presents a benchmark dataset evaluating unified page-level document parsing alongside alternative text descriptions for visual elements. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Introduces a document parsing dataset rather than LLM-wide evaluation.
Doc2Harness: Benchmarking LLMs on Long-Horizon Harness Engineering of Coding Agents The paper introduces a benchmark evaluating how well language models engineer isolated execution harnesses for coding agents using specification documents. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Evaluates software engineering capability, not safety or guardrail robustness.
Non-asymptotic Convergence of Average-reward Q-learning with Options The paper provides finite-sample convergence guarantees for average-reward option Q-learning algorithms under inter- and intra-option settings. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Theoretical RL analysis, not deployment safety or alignment robustness.
Non-asymptotic Convergence of Average-reward Q-learning with Options The paper provides finite-sample convergence guarantees for average-reward option Q-learning algorithms under inter- and intra-option settings. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on classical RL convergence rates, not efficient serving.
Non-asymptotic Convergence of Average-reward Q-learning with Options The paper provides finite-sample convergence guarantees for average-reward option Q-learning algorithms under inter- and intra-option settings. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Presents theoretical RL convergence proofs, not LLM agent systems.
Non-asymptotic Convergence of Average-reward Q-learning with Options The paper provides finite-sample convergence guarantees for average-reward option Q-learning algorithms under inter- and intra-option settings. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Contains theoretical convergence bounds, not LLM benchmark evaluation.
Lifted State Policy Optimization: Internalizing Self-Correction through Transition Geometry The paper proposes a policy optimization method that treats reasoning transitions as learned coordinates in an energy space to internalize self-correction. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on reasoning accuracy and latency, not deployment guardrails.
Lifted State Policy Optimization: Internalizing Self-Correction through Transition Geometry The paper proposes a policy optimization method that treats reasoning transitions as learned coordinates in an energy space to internalize self-correction. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes an optimization algorithm, not evaluation datasets or benchmarks.
Human Demonstrations as Post-Training Data for Any Robot Body: Retargeting the Object, Not the Hand The paper retargets object motion rather than human hands from video demonstrations to generate post-training data for various robot embodiments. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Presents a robot retargeting method rather than benchmark evaluation.
Human Demonstrations as Post-Training Data for Any Robot Body: Retargeting the Object, Not the Hand The paper retargets object motion rather than human hands from video demonstrations to generate post-training data for various robot embodiments. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Focuses on robotics embodiment transfer rather than LLM-driven planning agents.
Human Demonstrations as Post-Training Data for Any Robot Body: Retargeting the Object, Not the Hand The paper retargets object motion rather than human hands from video demonstrations to generate post-training data for various robot embodiments. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Studies general robotic manipulation transfer, not industrial vision defect inspection.
Human Demonstrations as Post-Training Data for Any Robot Body: Retargeting the Object, Not the Hand The paper retargets object motion rather than human hands from video demonstrations to generate post-training data for various robot embodiments. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Addresses motion transfer feasibility, not AI safety or jailbreaks.
From Relevance to Validity: Zero-Shot Tool Selection via Functional and Contextual Validation The paper proposes a progressive validation framework for zero-shot tool selection that evaluates functional compatibility and contextual relevance. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on agent tool selection accuracy, not deployment safety.
From Relevance to Validity: Zero-Shot Tool Selection via Functional and Contextual Validation The paper proposes a progressive validation framework for zero-shot tool selection that evaluates functional compatibility and contextual relevance. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes a tool selection framework rather than an evaluation benchmark.
Cortex: Consolidating Context into Parametric Memory The paper presents a method to consolidate context gains into parametric model weights using prospective queries and low-rank feed-forward updates. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Focuses on knowledge retention and editing, not safety guardrails.
Cortex: Consolidating Context into Parametric Memory The paper presents a method to consolidate context gains into parametric model weights using prospective queries and low-rank feed-forward updates. Preprint	Multimodal models via arxiv · 15 Oct, 15:06	Not relevant Evaluates general context consolidation across backbones, not core multimodal capability.
Cortex: Consolidating Context into Parametric Memory The paper presents a method to consolidate context gains into parametric model weights using prospective queries and low-rank feed-forward updates. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Targets model memory updating rather than autonomous agent planning systems.
Cortex: Consolidating Context into Parametric Memory The paper presents a method to consolidate context gains into parametric model weights using prospective queries and low-rank feed-forward updates. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on parametric context retention rather than small/efficient model architectures.
Cortex: Consolidating Context into Parametric Memory The paper presents a method to consolidate context gains into parametric model weights using prospective queries and low-rank feed-forward updates. Preprint	LLMs for code via arxiv · 15 Oct, 15:06	Not relevant Addresses general context consolidation rather than software engineering tasks.
Cortex: Consolidating Context into Parametric Memory The paper presents a method to consolidate context gains into parametric model weights using prospective queries and low-rank feed-forward updates. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Proposes a memory consolidation method, not general LLM benchmarking.
Cortex: Consolidating Context into Parametric Memory The paper presents a method to consolidate context gains into parametric model weights using prospective queries and low-rank feed-forward updates. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 15:06	Not relevant Not focused on industrial defect inspection or manufacturing quality.
When Does Momentum Help in Reinforcement Learning with Verifiable Rewards? The paper analyzes when first-moment momentum accelerates reinforcement learning with verifiable rewards under exact and sampled policy gradients. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Analyzes training optimization dynamics rather than benchmark evaluation methodologies.
When Does Momentum Help in Reinforcement Learning with Verifiable Rewards? The paper analyzes when first-moment momentum accelerates reinforcement learning with verifiable rewards under exact and sampled policy gradients. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Investigates optimization convergence rates, not safety alignment or guardrails.
When Does Momentum Help in Reinforcement Learning with Verifiable Rewards? The paper analyzes when first-moment momentum accelerates reinforcement learning with verifiable rewards under exact and sampled policy gradients. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Studies optimization theory for RL post-training, not agent workflows.
When Does Momentum Help in Reinforcement Learning with Verifiable Rewards? The paper analyzes when first-moment momentum accelerates reinforcement learning with verifiable rewards under exact and sampled policy gradients. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Focuses on optimizer acceleration rather than model compression or inference.
One Size Does Not Fit All: Curvature-Aware Adaptive Perturbation for SAM The paper introduces an adaptive perturbation method for sharpness-aware minimization that modulates perturbation radii using local loss curvature. Preprint	AI agents via arxiv · 15 Oct, 15:06	Not relevant Presents a generic neural network optimizer, not AI agent systems.
One Size Does Not Fit All: Curvature-Aware Adaptive Perturbation for SAM The paper introduces an adaptive perturbation method for sharpness-aware minimization that modulates perturbation radii using local loss curvature. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Targets general loss landscape optimization, not model compression or serving.
One Size Does Not Fit All: Curvature-Aware Adaptive Perturbation for SAM The paper introduces an adaptive perturbation method for sharpness-aware minimization that modulates perturbation radii using local loss curvature. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 15:06	Not relevant Studies general neural network loss curvature, not LLM safety guardrails.
One Size Does Not Fit All: Curvature-Aware Adaptive Perturbation for SAM The paper introduces an adaptive perturbation method for sharpness-aware minimization that modulates perturbation radii using local loss curvature. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06	Not relevant Focuses on an optimization algorithm, not LLM benchmark evaluation.
From Unreliable to Usable: Decomposing Failure Modes and Confidence Signals in LLM Entity Extraction - Evidence from Cyber Threat Intelligence The paper evaluates failure modes and confidence calibration across frontier and lightweight language models during entity extraction from cyber threat intelligence reports. Preprint	Small and efficient models via arxiv · 15 Oct, 15:06	Not relevant Evaluates existing low-cost models rather than proposing efficiency techniques.
From Unreliable to Usable: Decomposing Failure Modes and Confidence Signals in LLM Entity Extraction - Evidence from Cyber Threat Intelligence The paper evaluates failure modes and confidence calibration across frontier and lightweight language models during entity extraction from cyber threat intelligence reports. Preprint	LLMs for code via arxiv · 15 Oct, 15:06	Not relevant Analyzes cyber threat entity extraction, not software code tasks.

7,981–8,040 of 17,313← Previous Next →