4/23/2026
Evian: Towards Explainable Visual Instruction-tuning Data Auditing
Evian: Explainable auditing framework for vision-language model training data, identifying nuanced semantic flaws and logical errors.
4/23/2026
Enhancing Research Idea Generation through Combinatorial Innovation and Multi-Agent Iterative Search Strategies
Multi-agent LLM system for research idea generation using combinatorial innovation and iterative search to reduce repetition and increase novelty.
4/23/2026
Toward Cross-Lingual Quality Classifiers for Multilingual Pretraining Data Selection
Cross-lingual quality classifiers for multilingual pretraining data selection, leveraging embedding space consistency across languages.
4/23/2026
LayerTracer: A Joint Task-Particle and Vulnerable-Layer Analysis framework for Arbitrary Large Language Model Architectures
LayerTracer: Framework for analyzing hierarchical representations and robustness across diverse LLM architectures (Transformer, Mamba, GateDeltaNet).
4/23/2026
Trust, Lies, and Long Memories: Emergent Social Dynamics and Reputation in Multi-Round Avalon with LLM Agents
Study of emergent social dynamics in LLM agents playing multi-round deception game with memory retention, analyzing reputation formation.
4/23/2026
Beyond ZOH: Advanced Discretization Strategies for Vision Mamba
Comparison of discretization strategies for Vision Mamba state space models to improve temporal fidelity in dynamic visual environments.
4/23/2026
RSRCC: A Remote Sensing Regional Change Comprehension Benchmark Constructed via Retrieval-Augmented Best-of-N Ranking
RSRCC: Remote sensing benchmark with 126k questions for change detection and explanation in satellite imagery using vision-language models.
4/23/2026
GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifiable Process Supervision for Effective Reasoning
GRPO-VPS: Improves Group Relative Policy Optimization with verifiable process supervision for better reasoning in LLMs through refined credit assignment.
4/23/2026
The Expense of Seeing: Attaining Trustworthy Multimodal Reasoning Within the Monolithic Paradigm
Analysis of trustworthiness issues in Vision-Language Models, arguing current VLMs fail to faithfully synthesize multimodal data and proposing solutions.
4/23/2026
ORPHEAS: A Cross-Lingual Greek-English Embedding Model for Retrieval-Augmented Generation
ORPHEAS: Bilingual Greek-English embedding model optimized for retrieval-augmented generation, addressing morphological complexity and domain-specific terminology.
4/23/2026
A Field Guide to Decision Making
arXiv paper on decision-making frameworks augmented by machine intelligence for high-consequence scenarios.
4/23/2026
Storm Surge Modeling, Bias Correction, Graph Neural Networks, Graph Convolution Networks
arXiv paper: StormNet uses graph neural networks and convolution for spatio-temporal storm surge forecasting with bias correction.
4/23/2026
QuanForge: A Mutation Testing Framework for Quantum Neural Networks
arXiv paper: QuanForge mutation testing framework for quantum neural networks addressing stochastic factors and interpretability challenges.
4/23/2026
ONOTE: Benchmarking Omnimodal Notation Processing for Expert-level Music Intelligence
arXiv paper benchmarking omnimodal notation processing for music intelligence across auditory, visual, and symbolic domains.
4/23/2026
COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling
Data-centric framework using parameter-efficient fine-tuning for adapting LLMs to multiple languages with reduced interference.
4/23/2026
Exploiting LLM-as-a-Judge Disposition on Free Text Legal QA via Prompt Optimization
Study on prompt optimization and judge selection for LLM-as-a-Judge evaluations in legal question answering tasks.
4/23/2026
Supplement Generation Training for Enhancing Agentic Task Performance
Training method for adapting smaller LLMs to agentic tasks by generating supplemental text without retraining large models.
4/23/2026
Coverage, Not Averages: Semantic Stratification for Trustworthy Retrieval Evaluation
Semantic stratification method for improving retrieval evaluation in RAG systems by addressing bias in evaluation set construction.
4/23/2026
Working Memory Constraints Scaffold Learning in Transformers under Data Scarcity
Investigation of human-like working memory constraints in Transformer architectures for improved learning under data scarcity.
4/23/2026
Can "AI" Be a Doctor? A Study of Empathy, Readability, and Alignment in Clinical LLMs
Multidimensional evaluation of general-purpose and medical domain LLMs on clinical communication standards.
4/23/2026
OMIBench: Benchmarking Olympiad-Level Multi-Image Reasoning in Large Vision-Language Model
Benchmark for evaluating Olympiad-level multi-image reasoning in vision-language models with distributed evidence.
4/23/2026
Convergent Evolution: How Different Language Models Learn Similar Number Representations
Analysis of how different language model architectures learn similar periodic number representations in Fourier domain.
4/23/2026
AVISE: Framework for Evaluating the Security of AI Systems
Open-source framework for identifying vulnerabilities and evaluating security of AI systems across critical domains.
4/23/2026
LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals
Method for building person-specific generative agents using LLMs grounded in self-report data for general-purpose behavioral simulation.
4/23/2026
A Survey of Scaling in Large Language Model Reasoning
Survey examining scaling strategies for LLM reasoning including multi-agent collaboration and impact on model performance.
4/23/2026
Context Attribution with Multi-Armed Bandit Optimization
Novel framework using multi-armed bandit optimization to identify influential context segments in retrieval-augmented generation systems.
4/23/2026
Cognitive Kernel-Pro: A Framework for Deep Research Agents and Agent Foundation Models Training
Open-source framework for building general AI agents with deep research and autonomous capabilities without relying on proprietary APIs.
4/23/2026
AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite
AstaBench provides rigorous benchmarking suite for evaluating AI agents on scientific research tasks including literature review, experiments, and data analysis.
4/23/2026
FELA: A Multi-Agent Evolutionary System for Feature Engineering of Industrial Event Log Data
FELA applies multi-agent evolutionary system to automated feature engineering for industrial event logs, handling scale, dimensionality, and temporal complexity.
4/23/2026
Efficient Test-Time Scaling of Multi-Step Reasoning by Probing Internal States of Large Language Models
Improves test-time scaling of multi-step LLM reasoning by probing internal states to efficiently verify intermediate steps without expensive Process Reward Models.
4/23/2026
Same Content, Different Answers: Cross-Modal Inconsistency in MLLMs
REST benchmarks evaluate cross-modal inconsistency in MLLMs using samples with identical semantic content in image, text, and mixed modalities.
4/23/2026
Epistemic Constitutionalism Or: how to avoid coherence bias
Proposes epistemic constitution framework for regulating how LLMs form and express beliefs, addressing source attribution bias and implicit epistemic policies.
4/23/2026
LEAD: Breaking the No-Recovery Bottleneck in Long-Horizon Reasoning
LEAD addresses no-recovery bottleneck in long-horizon LLM reasoning by analyzing error distribution and showing balanced decomposition prevents irreversible failures on hard steps.
4/23/2026
Lightweight LLM Agent Memory with Small Language Models
Proposes lightweight external memory system for LLM agents using small language models, reducing overhead while maintaining accuracy for cross-turn consistency in long-horizon tasks.
4/23/2026
Model Space Reasoning as Search in Feedback Space for Planning Domain Generation
Uses agentic LLM feedback frameworks for planning domain generation from natural language, improving quality of generated domains through iterative model space reasoning.
4/23/2026
Seven simple steps for log analysis in AI systems
Proposes standardized pipeline for log analysis in AI systems with Python code examples, establishing best practices for understanding model capabilities and evaluation integrity.
4/23/2026
RAG-KT: Cross-platform Explainable Knowledge Tracing with Multi-view Fusion Retrieval Generation
RAG-KT applies LLM-based retrieval-augmented generation to knowledge tracing for student performance prediction, improving transfer and interpretability across learning platforms.
4/23/2026
QuarkMedSearch: A Long-Horizon Deep Search Agent for Exploring Medical Intelligence
QuarkMedSearch builds a long-horizon deep search agent for Chinese medical domain, systematically exploring data construction, training strategies, and evaluation for vertical agentic systems.
4/23/2026
TREX: Automating LLM Fine-tuning via Agent-Driven Tree-based Exploration
TREX automates LLM fine-tuning via multi-agent tree-based exploration, orchestrating researcher and executor agents to handle requirement analysis, experimentation, and optimization.
4/23/2026
MirrorBench: Evaluating Self-centric Intelligence in MLLMs by Introducing a Mirror
MirrorBench evaluates self-centric intelligence in MLLMs by introducing mirror reflection tasks, addressing gap in embodied AI benchmarks focused on external object understanding.
4/23/2026
Hybrid Decision Making via Conformal VLM-generated Guidance
Applies conformal prediction to VLM-generated guidance in hybrid decision-making, ensuring reliable human-AI collaboration where AI provides guidance rather than decisions.
4/23/2026
PersonalHomeBench: Evaluating Agents in Personalized Smart Homes
PersonalHomeBench evaluates foundation models as agentic assistants in personalized smart home environments through iterative benchmark construction with rich household states.
4/23/2026
Beyond Text-Dominance: Understanding Modality Preference of Omni-modal Large Language Models
Analyzes modality preference bias in omni-modal LLMs using a conflict-based benchmark, revealing that native unified architectures still exhibit systematic modality preference patterns.
4/23/2026
From Admission to Invariants: Measuring Deviation in Delegated Agent Systems
Studies structural limits of enforcement mechanisms in autonomous agent systems, showing how behavioral drift can become invisible to enforcement engines operating at wrong abstraction levels.
4/23/2026
CoSearch: Joint Training of Reasoning and Document Ranking via Reinforcement Learning for Agentic Search
CoSearch jointly trains reasoning and document ranking via RL for agentic search, optimizing both agent reasoning and retrieval components together rather than treating retrieval as fixed.
4/23/2026
LiteResearcher: A Scalable Agentic RL Training Framework for Deep Research Agent
LiteResearcher proposes scalable RL framework for training deep research agents, decoupling real-world search from training.
4/23/2026
Explicit Trait Inference for Multi-Agent Coordination
Explicit Trait Inference improves LLM-based multi-agent coordination by tracking partner warmth and competence characteristics.
4/23/2026
Fairness Testing of Large Language Models in Role-Playing
Studies fairness and bias in LLMs during role-playing scenarios, assessing whether social biases emerge in simulated roles.
4/23/2026
BatchLLM: Optimizing Large Batched LLM Inference with Global Prefix Sharing and Throughput-oriented Token Batching
BatchLLM optimizes batched LLM inference throughput using global prefix sharing and throughput-oriented token batching techniques.
4/23/2026
Recency Biased Causal Attention for Time-series Forecasting
Proposes recency bias mechanism for transformer attention in time-series forecasting using attention score reweighting.