2/24/2026
Neurosymbolic Retrievers for Retrieval-augmented Generation
Neurosymbolic retrievers for RAG combining neural and symbolic reasoning to improve interpretability and reduce hallucination.
2/24/2026
The Illusion of Human AI Parity Under Uncertainty: Navigating Elusive Ground Truth via a Probabilistic Paradigm
Probabilistic framework for benchmarking AI system capabilities accounting for uncertainty in ground truth expert judgments.
2/24/2026
Autonomous Business System via Neuro-symbolic AI
Neuro-symbolic AI system combining LLMs with deterministic logic for autonomous business process reconfiguration and cross-functional automation.
2/24/2026
OffSeeker: Online Reinforcement Learning Is Not All You Need for Deep Research Agents
OffSeeker demonstrates offline reinforcement learning for research agents reduces cost versus online RL while maintaining performance.
2/24/2026
MemOCR: Layout-Aware Visual Memory for Efficient Long-Horizon Reasoning
MemOCR: multimodal memory system for long-horizon agent reasoning that compresses interaction histories with layout-aware visual memory.
2/24/2026
Open Problems in Differentiable Social Choice: Learning Mechanisms, Decisions, and Alignment
Framework identifying open problems in differentiable social choice mechanisms for ML systems including auctions, resource allocation, and LLM alignment.
2/24/2026
SweetSpot: An Analytical Model for Predicting Energy Efficiency of LLM Inference
SweetSpot analytical model predicts energy efficiency of LLM inference by modeling non-linear relationships in autoregressive Transformer structures.
2/24/2026
ToolSelf: Unifying Task Execution and Self-Reconfiguration via Tool-Driven Intrinsic Adaptation
ToolSelf framework enables LLM agents to self-reconfigure dynamically during task execution via tool-driven intrinsic adaptation beyond static agent configurations.
2/24/2026
Interpretable Failure Analysis in Multi-Agent Reinforcement Learning Systems
Gradient-based framework for interpretable failure detection and attribution in multi-agent RL systems including patient-zero identification and domino effects.
2/24/2026
Does Your Reasoning Model Implicitly Know When to Stop Thinking?
Studies whether reasoning models implicitly learn when to stop reasoning, addressing inefficiency and redundancy in long chain-of-thought outputs.
2/24/2026
TSR: Trajectory-Search Rollouts for Multi-Turn RL of LLM Agents
TSR algorithm for multi-turn RL training of LLM agents addressing sparse/delayed rewards and stochastic environments through trajectory-search rollouts.
2/24/2026
CM2: Reinforcement Learning with Checklist Rewards for Multi-Turn and Multi-Step Agentic Tool Use
CM2 applies RL with checklist rewards to multi-turn, multi-step agentic tool use for open-ended objectives without fully verifiable reward functions.
2/24/2026
REMem: Reasoning with Episodic Memory in Language Agent
REMem framework enables language agents to perform reasoning using episodic memory with spatiotemporal context from interaction histories, not just semantic memory.
2/24/2026
ForesightSafety Bench: A Frontier Risk Evaluation and Governance Framework towards Safe AI
ForesightSafety Bench framework for evaluating frontier risks in autonomous AI systems addressing limitations of current safety benchmarks and alignment technologies.
2/24/2026
Competition for attention predicts good-to-bad tipping in AI
Analyzes attention competition in local LLMs as predictor for dangerous behavior tipping without cloud connectivity or explicit oversight mechanisms.
2/24/2026
Unifying Evolutionary Prompt Search and Reinforcement Learning for LLM Self-Improvement
E-SPL method jointly improves LLM context and weights by combining evolutionary prompt search with reinforcement learning for autonomous self-improvement.
2/24/2026
EnterpriseBench Corecraft: Training Generalizable Agents on High-Fidelity RL Environments
CoreCraft environment in EnterpriseBench suite trains AI agents on high-fidelity RL simulation of enterprise customer support with 2,500+ entities and 23 tools.
2/24/2026
Toward Scalable Verifiable Reward: Proxy State-Based Evaluation for Multi-turn Tool-Calling LLM Agents
Proposes Proxy State-Based Evaluation using LLM-driven simulation for benchmarking multi-turn tool-calling agents with scalable verifiable rewards.
2/24/2026
Towards a Science of AI Agent Reliability
Framework for evaluating AI agent reliability beyond accuracy metrics, examining consistency across runs and robustness to perturbations in deployed agents.
2/24/2026
LLM-WikiRace Benchmark: How Far Can LLMs Plan over Real-World Knowledge Graphs?
LLM-WikiRace benchmark evaluates LLM planning, reasoning, and world knowledge by requiring models to navigate Wikipedia hyperlinks to reach target pages efficiently.
2/24/2026
RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models
Introduces RFEval framework for evaluating reasoning faithfulness in large reasoning models using stance consistency and causal influence under counterfactual interventions.
2/24/2026
Dataless Weight Disentanglement in Task Arithmetic via Kronecker-Factored Approximate Curvature
Proposes Kronecker-factored approach for disentangling task vectors in task arithmetic without external data, reducing cross-task interference.
2/24/2026
ODESteer: A Unified ODE-Based Steering Framework for LLM Alignment
Introduces ODESteer, a unified ODE-based framework for LLM alignment via activation steering that captures complex activation distributions.
2/24/2026
Resource-Aware Distributed Submodular Maximization: A Paradigm for Multi-Robot Decision-Making
Proposes distributed submodular maximization algorithm for multi-robot decision-making balancing resource constraints and coordination complexity.
2/24/2026
Layer Collapse Can be Induced by Unstructured Pruning
Analyzes conditions under which unstructured pruning can induce layer collapse and structural effects in rectifier-activated networks.
2/24/2026
Adaptive Runge-Kutta Dynamics for Spatiotemporal Prediction
Proposes adaptive Runge-Kutta dynamics for spatiotemporal prediction incorporating physical knowledge for weather forecasting and video processing.
2/24/2026
(PASS) Visual Prompt Locates Good Structure Sparsity through a Recurrent HyperNetwork
Introduces PASS, a method using visual prompts with recurrent hypernetworks to identify effective structural sparsity patterns for neural network compression.
2/24/2026
Improving Discrete Optimisation Via Decoupled Straight-Through Estimator
Proposes decoupled straight-through estimator for improving gradient-based optimization of neural networks with discrete variables.
2/24/2026
Uncovering Autoregressive LLM Knowledge of Thematic Fit in Event Representation
Investigates LLM knowledge of thematic fit in semantic role prediction through prompt design experiments, achieving SOTA on benchmarks.
2/24/2026
Robust Time Series Causal Discovery for Agent-Based Model Validation
Proposes robust causal discovery method for validating agent-based models using time series data with improved accuracy on noisy data.
2/24/2026
Federated Co-tuning Framework for Large and Small Language Models
Presents FedCoLLM, a federated co-tuning framework enabling mutual enhancement between large and small language models with parameter efficiency.
2/24/2026
PoTable: Towards Systematic Thinking via Plan-then-Execute Stage Reasoning on Tables
Proposes PoTable, a plan-then-execute reasoning framework for improving LLM performance on table reasoning tasks.
2/24/2026
Evaluating LLMs' Divergent Thinking Capabilities for Scientific Idea Generation with Minimal Context
Introduces LiveIdeaBench, a benchmark for evaluating LLMs' scientific idea generation and divergent thinking with minimal context.
2/24/2026
Revisiting Graph Neural Networks for Graph-level Tasks: Taxonomy, Empirical Study, and Future Directions
Comprehensive empirical study and taxonomy of Graph Neural Networks for graph-level prediction tasks with analysis of evaluation methodologies.
2/24/2026
Dialogue is Better Than Monologue: Instructing Medical LLMs via Strategical Conversations
Introduces benchmark for evaluating medical LLMs using dialogue-based diagnostic scenarios with noise and difficulty levels instead of static QA tasks.
2/24/2026
The Curse of Depth in Large Language Models
Research identifying 'Curse of Depth' phenomenon where ~50% of LLM layers underperform. Analysis across Llama, Mistral, DeepSeek, Qwen with theoretical and empirical investigation.
2/24/2026
Analysis of Off-Policy $n$-Step TD-Learning with Linear Function Approximation
Convergence analysis of multi-step temporal difference learning with linear function approximation.
2/24/2026
From Contextual Combinatorial Semi-Bandits to Bandit List Classification: Improved Sample Complexity with Sparse Rewards
Theoretical analysis of contextual combinatorial bandits with sparse rewards for recommendation systems.
2/24/2026
VQEL: Enabling Self-Play in Emergent Language Games via Agent-Internal Vector Quantization
Vector quantization approach for emergent language learning enabling self-play among communicating agents.
2/24/2026
VideoMind: A Chain-of-LoRA Agent for Temporal-Grounded Video Reasoning
LLM agent using LoRA adapters for temporal-grounded video reasoning with interpretable evidence links.
2/24/2026
Learn by Reasoning: Analogical Weight Generation for Few-Shot Class-Incremental Learning
Few-shot class-incremental learning using analogical reasoning for weight generation of new classes.
2/24/2026
Beyond Single-Turn: A Survey on Multi-Turn Interactions with Large Language Models
Survey of multi-turn LLM interactions covering instruction following, reasoning, tool use, and conversational understanding.
2/24/2026
Towards A Universal Graph Structural Encoder
Pre-training approach for learning universal graph structural representations across different graph domains.
2/24/2026
GRILL: Restoring Gradient Signal in Ill-Conditioned Layers for More Effective Adversarial Attacks on Autoencoders
Attack method against autoencoders using gradient restoration to find more effective adversarial perturbations.
2/24/2026
Feature Representation Transferring to Lightweight Models via Perception Coherence
Knowledge distillation method using perception coherence loss to transfer representations to lightweight models.
2/24/2026
CodePDE: An Inference Framework for LLM-driven PDE Solver Generation
Framework leveraging LLMs for code generation to solve PDEs, combining symbolic reasoning with neural approaches.
2/24/2026
FairSHAP: Preprocessing for Fairness Through Attribution-Based Data Augmentation
Fairness preprocessing framework using SHAP values for transparent data augmentation to mitigate model bias.
2/24/2026
Heterogeneity-Aware Client Sampling for Optimal and Efficient Federated Learning
Client sampling strategy for federated learning addressing communication and computational heterogeneity.
2/24/2026
BitHydra: Towards Bit-flip Inference Cost Attack against Large Language Models
Attack method against LLMs using bit-flip manipulation of model parameters to maximize inference costs.
2/24/2026
Accidental Vulnerability: Factors in Fine-Tuning that Shift Model Safeguards
Investigation of vulnerabilities accidentally introduced during LLM fine-tuning on domain-specific datasets.