2/24/2026
The Problem of Algorithmic Collisions: Mitigating Unforeseen Risks in a Connected World
Analysis of systemic risks from interactions between deployed AI and algorithmic systems without mutual awareness.
2/24/2026
It Takes a Good Model to Train a Good Model: Generalized Gaussian Priors for Optimized LLMs
Study showing LLM weights follow generalized Gaussian distributions, proposing optimization framework for improved training efficiency and initialization.
2/24/2026
Cross-lingual Collapse: How Language-Centric Foundation Models Shape Reasoning in Large Language Models
Studies cross-lingual collapse phenomenon where multilingual LLMs revert to dominant language during reinforcement learning with verifiable rewards training.
2/24/2026
AbstRaL: Augmenting LLMs' Reasoning by Reinforcing Abstract Thinking
AbstRaL method improves LLM reasoning on grade school math through synthetic data generation to enhance robustness against distribution shifts.
2/24/2026
Can Generalist Vision Language Models (VLMs) Rival Specialist Medical VLMs? Benchmarking and Strategic Insights
Benchmark comparing generalist vs specialist medical vision-language models for clinical image diagnosis.
2/24/2026
Modular Delta Merging with Orthogonal Constraints: A Scalable Framework for Continual and Reversible Model Composition
MDM-OC framework for scalable, reversible model composition enabling continual learning without catastrophic forgetting.
2/24/2026
Role-Aware Language Models for Secure and Contextualized Access Control in Organizations
Fine-tuning approach enabling LLMs to enforce role-based access control and generate contextualized responses in enterprise settings.
2/24/2026
Generative Logic: A New Computer Architecture for Deterministic Reasoning and Knowledge Generation
Generative Logic deterministic architecture for deductive reasoning from axioms via Mathematical Programming Language.
2/24/2026
MolReasoner: Toward Effective and Interpretable Reasoning for Molecular LLMs
MolReasoner framework for domain-specific molecular reasoning in LLMs using specialized prompting to reduce hallucinations.
2/24/2026
Shuffle-R1: Efficient RL framework for Multimodal Large Language Models via Data-centric Dynamic Shuffle
Shuffle-R1 framework addressing training inefficiencies in reinforcement learning for multimodal LLMs through data-centric dynamic shuffling.
2/24/2026
Benchmarking Pretrained Molecular Embedding Models For Molecular Representation Learning
Comprehensive benchmark comparing 25 pretrained molecular embedding models across 25 datasets for chemistry applications.
2/24/2026
CORE: Measuring Multi-Agent LLM Interaction Quality under Game-Theoretic Pressures
CORE metric for evaluating linguistic effectiveness in multi-agent LLM interactions under game-theoretic pressures.
2/24/2026
AFABench: A Generic Framework for Benchmarking Active Feature Acquisition
Benchmark framework for active feature acquisition methods that dynamically select informative features under cost constraints.
2/24/2026
AttestLLM: Efficient Attestation Framework for Billion-scale On-device LLMs
Attestation framework for verifying legitimacy of billion-parameter LLMs running on-device with efficiency constraints.
2/24/2026
Lightweight error mitigation strategies for post-training N:M activation sparsity in LLMs
Lightweight error mitigation strategies for post-training N:M structured activation pruning in LLMs enabling dynamic input-adaptive compression.
2/24/2026
SpinGPT: A Large-Language-Model Approach to Playing Poker Correctly
LLM-based approach to poker using counterfactual regret minimization principles for multi-player game strategy beyond Nash equilibrium.
2/24/2026
Look Back to Reason Forward: Revisitable Memory for Long-Context LLM Agents
Revisitable memory system for long-context LLM agents enabling efficient retrieval and reasoning across millions of tokens via dynamic evidence management.
2/24/2026
Polychromic Objectives for Reinforcement Learning
Polychromic objectives method for reinforcement learning fine-tuning preventing policy collapse and preserving diversity during RLFT training.
2/24/2026
On Predictability of Reinforcement Learning Dynamics for Large Language Models
Analysis of reinforcement learning parameter dynamics in LLMs identifying rank-1 dominance property and predictability patterns during reasoning improvements.
2/24/2026
The Unseen Frontier: Pushing the Limits of LLM Sparsity with Surrogate-Free ADMM
Surrogate-free ADMM pruning method achieving high sparsity levels (>60%) in LLMs without severe accuracy degradation, advancing neural network compression.
2/24/2026
RewardMap: Tackling Sparse Rewards in Fine-grained Visual Reasoning via Multi-Stage Reinforcement Learning
Multi-stage reinforcement learning approach (RewardMap) tackling sparse rewards in fine-grained visual reasoning tasks for multimodal LLMs.
2/24/2026
KVComm: Enabling Efficient LLM Communication through Selective KV Sharing
KVComm framework enabling efficient multi-agent LLM communication through selective key-value cache sharing, reducing inference costs versus natural language protocols.
2/24/2026
SocialHarmBench: Revealing LLM Vulnerabilities to Socially Harmful Requests
Safety benchmark dataset of 585 prompts across 7 sociopolitical categories in 34 countries testing LLM vulnerabilities to political manipulation, propaganda, and disinformation.
2/24/2026
AgentDR: Dynamic Recommendation with Implicit Item-Item Relations via LLM-based Agents
LLM-based agent framework for dynamic recommendations leveraging commonsense reasoning to capture implicit item-item relationships and avoid hallucination.
2/24/2026
EconCausal: A Context-Aware Causal Reasoning Benchmark for Large Language Models in Social Science
Context-aware causal reasoning benchmark for LLMs in social science, testing ability to distinguish structural causal mechanisms from correlations under varying institutional contexts.
2/24/2026
Towards Information-Optimized Multi-Agent Path Finding: A Hybrid Framework with Reduced Inter-Agent Information Sharing
Hybrid multi-agent pathfinding framework combining centralized and distributed approaches with reduced inter-agent information sharing for scalable autonomous systems.
2/24/2026
DITTO: A Spoofing Attack Framework on Watermarked LLMs via Knowledge Distillation
Attack framework demonstrating watermark spoofing vulnerabilities in LLMs via knowledge distillation, enabling malicious models to generate text with victim model watermarks.
2/24/2026
A Primer on SO(3) Action Representations in Deep Reinforcement Learning
Analysis of SO(3) action representations in deep reinforcement learning for robotic control, comparing Euler angles, quaternions, rotation matrices, and Lie algebra coordinates.
2/24/2026
Medical Interpretability and Knowledge Maps of Large Language Models
Systematic study of medical-domain interpretability in LLMs using UMAP projections, gradient-based saliency, layer lesioning, and activation patching to understand knowledge representation.
2/24/2026
Closing the Gap Between Text and Speech Understanding in LLMs
Analyzes performance gap between speech-adapted and text-based LLMs on language understanding tasks, identifying causes of speech input underperformance.
2/24/2026
From Pixels to Words -- Towards Native Vision-Language Primitives at Scale
Analysis of native Vision-Language Models architecture constraints versus modular approaches, discussing fundamental barriers and methods for accessible VLM research.
2/24/2026
Lean Finder: Semantic Search for Mathlib That Understands User Intents
Lean Finder: Semantic search engine for Lean and mathlib using intent-aware matching to help mathematicians locate relevant theorems and reduce learning curve.
2/24/2026
MoMaGen: Generating Demonstrations under Soft and Hard Constraints for Multi-Step Bimanual Mobile Manipulation
MoMaGen: Automated demonstration generation framework for training robots on multi-step bimanual mobile manipulation tasks under soft and hard constraints.
2/24/2026
Transitive RL: Value Learning via Divide and Conquer
Transitive RL: Divide-and-conquer value learning algorithm for offline goal-conditioned reinforcement learning using triangle inequality structure.
2/24/2026
Incentivizing Agentic Reasoning in LLM Judges via Tool-Integrated Reinforcement Learning
TIR-Judge: RL framework training LLM judges to use tools for constraint verification and computation, improving evaluation beyond text-based reasoning alone.
2/24/2026
Beyond a Million Tokens: Benchmarking and Enhancing Long-Term Memory in LLMs
Benchmark framework for evaluating long-context reasoning in LLMs beyond one million tokens with coherent narratives for conversational long-term memory assessment.
2/24/2026
PoCo: Agentic Proof-of-Concept Exploit Generation for Smart Contracts
PoCo: Agentic system for automated proof-of-concept exploit generation in smart contract security audits using AI-driven vulnerability demonstration.
2/24/2026
Scalable Multi-Objective and Meta Reinforcement Learning via Gradient Estimation
Scalable multi-objective and meta-RL via gradient estimation: efficient policy training across multiple related objectives through optimal task grouping.
2/24/2026
Agentic AI Systems in Electrical Power Systems Engineering: Current State-of-the-Art and Challenges
Comprehensive review establishing definitions and taxonomy of agentic AI systems in electrical power engineering, differentiating from traditional AI agents and generative models.
2/24/2026
Mantis: A Versatile Vision-Language-Action Model with Disentangled Visual Foresight
Mantis: Vision-Language-Action model with disentangled visual foresight for improved efficiency in VLA training and action prediction from visual input.
2/24/2026
InTAct: Interval-based Task Activation Consolidation for Continual Learning
InTAct: Continual learning approach using interval-based task activation consolidation with mathematical guarantees against catastrophic forgetting in neural networks.
2/24/2026
From Competition to Coordination: Market Making as a Scalable Framework for Safe and Aligned Multi-Agent LLM Systems
Market-making framework for coordinating multi-agent LLM systems, replacing centralized oversight with scalable coordination mechanisms for transparency and accountability.
2/24/2026
FAST: Topology-Aware Frequency-Domain Distribution Matching for Coreset Selection
FAST: Topology-aware coreset selection method for compressing datasets while maintaining distributional equivalence, reducing computational burden for neural network training.
2/24/2026
MapReduce LoRA: Advancing the Pareto Front in Multi-Preference Optimization for Generative Models
MapReduce LoRA and RaTE methods address multi-objective optimization in generative models, training preference-specific adapters to improve RLHF alignment without degrading other dimensions.
2/24/2026
SelfAI: A self-directed framework for long-horizon scientific discovery
SelfAI: Multi-agent system for long-horizon scientific discovery with self-directed exploration, balancing efficiency-diversity trade-offs in complex hypothesis spaces.
2/24/2026
ML-Tool-Bench: Tool-Augmented Planning for ML Tasks
ML-Tool-Bench: Framework for autonomous ML agents using LLMs to orchestrate end-to-end data science workflows including analysis, feature engineering, and hyperparameter optimization.
2/24/2026
Setting up for failure: automatic discovery of the neural mechanisms of cognitive errors
Automated discovery of RNN mechanisms for understanding cognitive errors and neural dynamics underlying behavior, replacing human-in-the-loop iterative refinement.
2/24/2026
DL$^3$M: A Vision-to-Language Framework for Expert-Level Medical Reasoning through Deep Learning and Large Language Models
Vision-to-language framework combining medical image classification with LLM reasoning for interpretable clinical decision support.
2/24/2026
Ev-Trust: An Evolutionary Stable Trust Mechanism for Decentralized LLM-Based Multi-Agent Service Economies
Trust mechanism for decentralized multi-agent LLM systems to prevent deceptive behavior and maintain system stability in autonomous service interactions.
2/24/2026
Clust-PSI-PFL: A Population Stability Index Approach for Clustered Non-IID Personalized Federated Learning
Clustered personalized federated learning framework using Population Stability Index to handle non-IID data distribution across clients.