2/24/2026
[b]=[d]-[t]+[p]: Self-supervised Speech Models Discover Phonological Vector Arithmetic
Analysis of self-supervised speech models across 96 languages showing phonological vector arithmetic structure.
2/24/2026
Early Evidence of Vibe-Proving with Consumer LLMs: A Case Study on Spectral Region Characterization with ChatGPT-5.2 (Thinking)
Case study using ChatGPT-5 thinking mode to resolve mathematical conjecture on spectral regions.
2/24/2026
Why Agent Caching Fails and How to Fix It: Structured Intent Canonicalization with Few-Shot Learning
Analysis of why LLM caching methods fail and proposes intent canonicalization with few-shot learning for cost reduction.
2/24/2026
Implicit Bias and Convergence of Matrix Stochastic Mirror Descent
Convergence analysis of Stochastic Mirror Descent with matrix parameters in overparameterized regime.
2/24/2026
Capable but Unreliable: Canonical Path Deviation as a Causal Mechanism of Agent Failure in Long-Horizon Tasks
Analysis of language agent failures on tool-use tasks caused by canonical path deviation rather than capability limitations.
2/24/2026
IAPO: Information-Aware Policy Optimization for Token-Efficient Reasoning
IAPO: information-theoretic post-training framework for optimizing token efficiency in LLM reasoning chains.
2/24/2026
Event-Triggered Gossip for Distributed Learning
Event-triggered gossip framework for distributed learning that reduces inter-node communication overhead.
2/24/2026
Beyond Behavioural Trade-Offs: Mechanistic Tracing of Pain-Pleasure Decisions in an LLM
Mechanistic interpretability study of how LLMs represent and use valence-related information in decision-making tasks.
2/24/2026
Characterizing MARL for Energy Control: A Multi-KPI Benchmark on the CityLearn Environment
Benchmark for Multi-Agent Reinforcement Learning algorithms on urban energy management tasks using CityLearn environment.
2/24/2026
Attention Deficits in Language Models: Causal Explanations for Procedural Hallucinations
Mechanistic analysis of procedural hallucinations in LLMs showing attention and readout-stage routing errors cause value retrieval failures.
2/24/2026
Scaling Laws for Precision in High-Dimensional Linear Regression
Theoretical study of how low-precision training affects scaling laws in high-dimensional linear regression with implications for quantization.
2/24/2026
Robust Exploration in Directed Controller Synthesis via Reinforcement Learning with Soft Mixture-of-Experts
Uses soft mixture-of-experts RL for exploration in directed controller synthesis with improved zero-shot generalization to larger systems.
2/24/2026
Prognostics of Multisensor Systems with Unknown and Unlabeled Failure Modes via Bayesian Nonparametric Process Mixtures
Bayesian nonparametric approach for predictive maintenance handling unknown failure modes in manufacturing without labeled data.
2/24/2026
TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics
TOPReward uses token probabilities from Vision-Language-Action models as zero-shot reward signals for robotics reinforcement learning without task-specific training.
2/24/2026
US-JEPA: A Joint Embedding Predictive Architecture for Medical Ultrasound
US-JEPA adapts Joint-Embedding Predictive Architectures for medical ultrasound by predicting masked latent representations instead of raw pixels.
2/24/2026
SplitLight: An Exploratory Toolkit for Recommender Systems Datasets and Splits
SplitLight open-source toolkit addresses reproducibility issues in recommender systems evaluation by exploring dataset preparation and splitting strategies.
2/24/2026
MentalBlackboard: Evaluating Spatial Visualization via Mathematical Transformations
MentalBlackboard benchmark evaluates Vision-Language Models on spatial visualization tasks like paper folding and hole punching.
2/24/2026
Seeing Farther and Smarter: Value-Guided Multi-Path Reflection for VLM Policy Optimization
Value-guided multi-path reflection method for optimizing Vision-Language Models in complex robotic manipulation through improved planning and reasoning.
2/24/2026
Adaptive Data Augmentation with Multi-armed Bandit: Sample-Efficient Embedding Calibration for Implicit Pattern Recognition
Multi-armed bandit approach for adaptive data augmentation to improve implicit pattern recognition in vision and language models.
2/24/2026
IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking
IR³ framework detects and mitigates reward hacking in RLHF by reverse-engineering and interpreting internalized objectives in LLM alignment.
2/24/2026
OptiRepair: Closed-Loop Diagnosis and Repair of Supply Chain Optimization Models with LLM Agents
OptiRepair uses LLM agents to diagnose and repair infeasible supply chain optimization models through closed-loop diagnosis and domain-agnostic repair.
2/24/2026
Human-Guided Agentic AI for Multimodal Clinical Prediction: Lessons from the AgentDS Healthcare Benchmark
Human-guided agentic AI system for multimodal clinical prediction, combining autonomous workflows with domain expertise on AgentDS Healthcare benchmark.
2/24/2026
Pyramid MoA: A Probabilistic Framework for Cost-Optimized Anytime Inference
Hierarchical Mixture-of-Agents architecture using lightweight router for cost-optimized LLM inference trading off accuracy and computational expense.
2/24/2026
Ada-RS: Adaptive Rejection Sampling for Selective Thinking
Adaptive Rejection Sampling framework for selective reasoning in LLMs, reducing token waste on simple requests while maintaining chain-of-thought benefits.
2/24/2026
Cost-Aware Diffusion Active Search
Cost-aware active search algorithm for autonomous agents to balance exploration and exploitation with unknown target recovery.
2/24/2026
Beyond a Single Extractor: Re-thinking HTML-to-Text Extraction for LLM Pretraining
Investigates HTML-to-text extraction methods for LLM pretraining datasets, showing single extractors lead to suboptimal web data coverage.
2/24/2026
Workflow-Level Design Principles for Trustworthy GenAI in Automotive System Engineering
Design principles for integrating LLMs into automotive system engineering with focus on trustworthiness and verification in safety-critical pipelines.
2/24/2026
Denoising Particle Filters: Learning State Estimation with Single-Step Objectives
Proposes denoising particle filters for robot state estimation trained on single-step objectives rather than sequence unrolling.
2/24/2026
Personalized Longitudinal Medical Report Generation via Temporally-Aware Federated Adaptation
Federated learning approach for personalized longitudinal medical report generation respecting privacy with temporal dynamics.
2/24/2026
SkillOrchestra: Learning to Route Agents via Skill Transfer
SkillOrchestra system for routing across multiple AI agents via learned skill transfer, reducing routing collapse in multi-turn conversations.
2/24/2026
Universal Pose Pretraining for Generalizable Vision-Language-Action Policies
Vision-Language-Action models with universal pose pretraining for improved 3D state perception and embodied AI tasks.
2/24/2026
Exact Discrete Stochastic Simulation with Deep-Learning-Scale Gradient Optimization
Enables gradient-based learning for continuous-time Markov chains by decoupling simulation from differentiation for optimization.
2/24/2026
Meta-Learning and Meta-Reinforcement Learning - Tracing the Path towards DeepMind's Adaptive Agent
Survey on meta-learning and meta-reinforcement learning techniques for rapid adaptation to novel tasks with minimal data.
2/24/2026
Orthogonal Uplift Learning with Permutation-Invariant Representations for Combinatorial Treatments
Uplift learning method for estimating causal effects of combinatorial treatments using permutation-invariant representations.
2/24/2026
Gradient based Severity Labeling for Biomarker Classification in OCT
arXiv paper proposing gradient-based severity labeling for contrastive learning in medical image biomarker classification.
2/24/2026
RobPI: Robust Private Inference against Malicious Client
arXiv paper on private inference protocols resilient to malicious client behavior in ML deployment.
2/24/2026
Janus-Q: End-to-End Event-Driven Trading via Hierarchical-Gated Reward Modeling
arXiv paper on event-driven trading using LLMs and hierarchical-gated reward modeling with textual market signals.
2/24/2026
Beyond Mimicry: Toward Lifelong Adaptability in Imitation Learning
arXiv paper proposing lifelong adaptability in imitation learning to improve compositional generalization beyond memorization.
2/24/2026
Multivariate time-series forecasting of ASTRI-Horn monitoring data: A Normal Behavior Model
arXiv paper on normal behavior modeling for time-series forecasting of telescope monitoring data.
2/24/2026
FairFS: Addressing Deep Feature Selection Biases for Recommender System
arXiv paper on feature importance estimation and selection biases in large-scale recommender systems.
2/24/2026
Closing the gap in multimodal medical representation alignment
arXiv paper addressing modality gap in CLIP-based multimodal learning for medical image representation alignment.
2/24/2026
The Invisible Gorilla Effect in Out-of-distribution Detection
arXiv paper on out-of-distribution detection in deep neural networks and how regional focus affects OOD detection performance.
2/24/2026
Descent-Guided Policy Gradient for Scalable Cooperative Multi-Agent Learning
Research on descent-guided policy gradients for multi-agent reinforcement learning addressing cross-agent noise scaling issues.
2/24/2026
Align When They Want, Complement When They Need! Human-Centered Ensembles for Adaptive Human-AI Collaboration
Human-AI collaboration framework using adaptive ensembles that align with human judgment when beneficial and complement when needed.
2/24/2026
ReSyn: Autonomously Scaling Synthetic Environments for Reasoning Models
Autonomous scaling method for synthetic training environments using verifier-based reinforcement learning for reasoning language models.
2/24/2026
NanoKnow: How to Know What Your Language Model Knows
Investigation of LLM knowledge encoding using nanochat family with fully open pre-training data to understand where parametric knowledge originates.
2/24/2026
JUCAL: Jointly Calibrating Aleatoric and Epistemic Uncertainty in Classification Tasks
Post-calibration uncertainty quantification method balancing aleatoric and epistemic uncertainty in classifier ensembles.
2/24/2026
Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks
Security vulnerability analysis of LLM agents exploiting skill files for prompt injection attacks through the agent skills supply chain.
2/24/2026
A Very Big Video Reasoning Suite
Benchmark suite for evaluating video reasoning capabilities of models, focusing on spatiotemporal consistency and reasoning beyond visual quality.
2/24/2026
A simple connection from loss flatness to compressed neural representations
Investigation of loss surface sharpness relationship to neural representation compression through local volumetric ratio and activation concentration measures.