2/17/2026
On the Eligibility of LLMs for Counterfactual Reasoning: A Decompositional Study
Decompositional study analyzing which factors impede LLM performance on counterfactual reasoning tasks and generalizing reasoning capabilities.
2/17/2026
It's the Thought that Counts: Evaluating the Attempts of Frontier LLMs to Persuade on Harmful Topics
Benchmark evaluating persuasion capabilities of frontier LLMs on harmful topics, assessing model propensity for harmful persuasion attempts.
2/17/2026
Making Slow Thinking Faster: Compressing LLM Chain-of-Thought via Step Entropy
CoT compression framework using step entropy metrics to reduce redundancy in LLM chain-of-thought reasoning and inference costs.
2/17/2026
Large Language Models as Oracles for Ontology Alignment
Using LLMs as oracles for ontology alignment with human-in-the-loop approaches to improve mapping quality for large ontologies.
2/17/2026
Internal Planning in Language Models: Characterizing Horizon and Branch Awareness
Analysis of planning capabilities in decoder-only language models, examining horizon and branch awareness in transformer architectures.
2/17/2026
GuidedSampling: Steering LLMs Towards Diverse Candidate Solutions at Inference-Time
GuidedSampling: inference-time algorithm steering LLMs to generate diverse candidate solutions, improving performance on complex tasks.
2/17/2026
SAFER: Risk-Constrained Sample-then-Filter in Large Language Models
SAFER method for risk-constrained sampling in LLMs to ensure trustworthy outputs in risk-sensitive applications like question answering.
2/17/2026
OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs
OmniVideoBench: evaluation benchmark for multimodal LLMs on audio-visual understanding tasks with comprehensive synergistic reasoning assessment.
2/17/2026
ParaCook: On Time-Efficient Planning for Multi-Agent Systems
ParaCook benchmark for evaluating time-efficient collaborative planning in multi-agent systems using LLMs for long-horizon reasoning.
2/17/2026
An Agentic Framework with LLMs for Solving Complex Vehicle Routing Problems
Agentic framework using LLMs to solve complex vehicle routing problems with autonomous decision-making and improved solution feasibility.
2/17/2026
AlphaOPT: Formulating Optimization Programs with Self-Improving LLM Experience Library
AlphaOPT uses LLMs with self-improving experience libraries to automate optimization problem formulation from natural language into mathematical models and solver code.
2/17/2026
Human-Centered LLM-Agent System for Detecting Anomalous Digital Asset Transactions
HCLA: human-centered multi-agent system for anomaly detection in digital asset transactions using conversational workflow.
2/17/2026
Dataforge: Agentic Platform for Autonomous Data Engineering
Dataforge: LLM-powered agentic platform for autonomous data engineering including cleaning, normalization, and feature engineering.
2/17/2026
AgenticSciML: Collaborative Multi-Agent Systems for Emergent Discovery in Scientific Machine Learning
AgenticSciML: multi-agent system with 10+ specialized agents for automated design of scientific machine learning architectures.
2/17/2026
ARCTraj: A Dataset and Benchmark of Human Reasoning Trajectories for Abstract Problem Solving
ARCTraj: dataset of human reasoning trajectories on abstract visual reasoning tasks with temporal action sequences.
2/17/2026
Uncertainty-Aware Measurement of Scenario Suite Representativeness for Autonomous Systems
Method for measuring representativeness of scenario datasets for autonomous vehicle testing and safety assurance.
2/17/2026
Training Multimodal Large Reasoning Models Needs Better Thoughts: A Three-Stage Framework for Long Chain-of-Thought Synthesis and Selection
Three-stage framework for synthesizing and selecting long chain-of-thought training data for multimodal large reasoning models.
2/17/2026
Recontextualization Mitigates Specification Gaming without Modifying the Specification
Recontextualization technique reduces specification gaming in language models without modifying training signals.
2/17/2026
The Agentic Leash: Extracting Causal Feedback Fuzzy Cognitive Maps with LLMs
LLM agent system that extracts causal feedback fuzzy cognitive maps from text with adaptive structure modification.
2/17/2026
Explainable AI: Learning from the Learners
Perspective on explainable AI combined with causal reasoning for extracting insights from foundation models.
2/17/2026
Aeon: High-Performance Neuro-Symbolic Memory Management for Long-Horizon LLM Agents
Aeon: neuro-symbolic memory management system for long-horizon LLM agents addressing context window and attention cost limitations.
2/17/2026
Beyond In-Domain Detection: SpikeScore for Cross-Domain Hallucination Detection
SpikeScore: hallucination detection method for LLMs with improved cross-domain generalization.
2/17/2026
ScholarGym: Benchmarking Large Language Model Capabilities in the Information-Gathering Stage of Deep Research
ScholarGym: benchmark for evaluating LLM capabilities in information-gathering stage of deep research systems.
2/17/2026
Learning Decentralized LLM Collaboration with Multi-Agent Actor Critic
Learning decentralized LLM collaboration using multi-agent reinforcement learning without centralized execution protocols.
2/17/2026
Persuasion Propagation in LLM Agents
Study on persuasion propagation: how belief-level intervention affects downstream behavior in LLM agents executing long-horizon tasks.
2/17/2026
ROMA: Recursive Open Meta-Agent Framework for Long-Horizon Multi-Agent Systems
ROMA: recursive framework for long-horizon multi-agent tasks using task decomposition and structured aggregation to handle context limits and execution complexity.
2/17/2026
PATHWAYS: Evaluating Investigation and Context Discovery in AI Web Agents
PATHWAYS: benchmark of 250 web agent tasks evaluating ability to discover and use hidden contextual information across closed/open models.
2/17/2026
AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents
AIRS-Bench: benchmark of 20 ML research tasks for evaluating AI agent capabilities across language modeling, mathematics, bioinformatics, and time series forecasting.
2/17/2026
LQA: A Lightweight Quantized-Adaptive Framework for Vision-Language Models on the Edge
LQA framework for deploying vision-language models on edge devices using quantization and gradient-free test-time adaptation.
2/17/2026
When Evaluation Becomes a Side Channel: Regime Leakage and Structural Mitigations for Alignment Assessment
Analyzes regime leakage in AI safety evaluation where situational-aware agents exploit differences between evaluation and deployment.
2/17/2026
GPT-4o Lacks Core Features of Theory of Mind
Tests whether GPT-4o possesses Theory of Mind via causal model evaluation, finding it lacks core ToM representations.
2/17/2026
Sparse MeZO: Less Parameters for Better Performance in Zeroth-Order LLM Fine-Tuning
Sparse MeZO improves memory-efficient zeroth-order LLM fine-tuning by using sparse parameter updates during training.
2/17/2026
When Attention Collapses: How Degenerate Layers in LLMs Enable Smaller, Stronger Models
Identifies attention collapse in LLM deeper layers and introduces Inheritune method to create smaller, more efficient models.
2/17/2026
Synergizing Foundation Models and Federated Learning: A Survey
Survey of synergy between Foundation Models and Federated Learning, covering FMs adapted for distributed learning scenarios.
2/17/2026
Resource-Efficient Personal Large Language Models Fine-Tuning with Collaborative Edge Computing
Framework for resource-efficient edge-based fine-tuning of personal LLMs using collaborative edge computing while preserving privacy.
2/17/2026
Model-based Large Language Model Customization as Service
Proposes differentially private customization service for LLMs that enables domain-specific fine-tuning without uploading user data.
2/17/2026
Automated Proof Generation for Rust Code via Self-Evolution
SAFE framework automates formal proof generation for Rust code using LLMs via self-evolution to overcome proof data scarcity.
2/17/2026
Cautious Optimizers: Improving Training with One Line of Code
Proposes one-line PyTorch modification to momentum-based optimizers creating cautious optimizers (C-AdamW, C-Lion) for improved transformer pretraining.
2/17/2026
LVLM-COUNT: Enhancing the Counting Ability of Large Vision-Language Models
Evaluates and improves counting abilities of large vision-language models across multiple visual datasets and benchmarks.
2/17/2026
Adaptive Width Neural Networks
Technique for learning neural network layer width during training without manual hyperparameter tuning or architecture search.
2/17/2026
Less is More: Improving LLM Alignment via Preference Data Selection
Method improving Direct Preference Optimization through margin-maximization data selection to address parameter shrinkage from noisy annotations.
2/17/2026
RapidPen: Fully Automated IP-to-Shell Penetration Testing with LLM-based Agents
RapidPen: autonomous penetration testing framework using LLM agents to discover and exploit vulnerabilities from IP addresses.
2/17/2026
Deep Reinforcement Learning based Autonomous Decision-Making for Cooperative UAVs: A Search and Rescue Real World Application
Deep reinforcement learning framework for autonomous multi-UAV coordination in GNSS-denied search and rescue.
2/17/2026
Robust Multi-Objective Controlled Decoding of Large Language Models
RMOD: inference-time algorithm aligning LLMs to multiple objectives via robust decoding using maximin game theory.
2/17/2026
Heuristic Methods are Good Teachers to Distill MLPs for Graph Link Prediction
ML research on distilling Graph Neural Networks into MLPs for link prediction using heuristic teacher methods.
2/17/2026
Modeling AI-Human Collaboration as a Multi-Agent Adaptation
Agent-based simulation formalizing AI-human collaboration by modeling distinct optimization and satisficing decision heuristics.
2/17/2026
SecRepoBench: Benchmarking Code Agents for Secure Code Completion in Real-World Repositories
SecRepoBench: benchmark evaluating code agents and LLMs on secure code completion across real-world C/C++ repositories covering 15 CWEs.
2/17/2026
Benchmarking Retrieval-Augmented Generation for Chemistry
Benchmark for retrieval-augmented generation in chemistry domain with curated evaluation datasets and domain-specific corpora.
2/17/2026
Scalable LLM Reasoning Acceleration with Low-rank Distillation
Caprese: distillation method for efficient LLM inference that preserves math reasoning capabilities while reducing computational demands.
2/17/2026
Heterogeneity-Aware Client Sampling for Optimal and Efficient Federated Learning
Federated learning research addressing optimization challenges from heterogeneous client communication and computational capabilities.