2/25/2026
UI-Venus-1.5 Technical Report
UI-Venus-1.5: GUI agent with 2B, 8B, and 30B-A3B variants for automating digital environment interactions with broad generality and strong task performance.
2/25/2026
VESPO: Variational Sequence-Level Soft Policy Optimization for Stable Off-Policy LLM Training
VESPO: reinforcement learning method for training LLMs with improved stability through soft policy optimization and importance sampling to address policy divergence.
2/25/2026
KBVQ-MoE: KLT-guided SVD with Bias-Corrected Vector Quantization for MoE Large Language Models
KBVQ-MoE: compression technique for Mixture of Experts LLMs using vector quantization and SVD to reduce parameter size and memory for resource-constrained deployment.
2/25/2026
Sim2Radar: Toward Bridging the Radar Sim-to-Real Gap with VLM-Guided Scene Reconstruction
Sim2Radar uses VLM-guided scene reconstruction to synthesize radar training data from RGB images, addressing radar dataset scarcity.
2/25/2026
Silent Inconsistency in Data-Parallel Full Fine-Tuning: Diagnosing Worker-Level Optimization Misalignment
Identifies and analyzes 'silent inconsistency' problem in data-parallel LLM fine-tuning where gradient synchronization doesn't ensure worker-level optimization alignment.
2/25/2026
ST-EVO: Towards Generative Spatio-Temporal Evolution of Multi-Agent Communication Topologies
ST-EVO framework for self-evolving LLM-powered multi-agent systems that dynamically construct task-adaptive communication topologies instead of predefined structures.
2/25/2026
Anatomy of Capability Emergence: Scale-Invariant Representation Collapse and Top-Down Reorganization in Neural Networks
Studies mechanical basis of capability emergence in neural networks across scales 405K-85M parameters, finding scale-invariant representation collapse and top-down reorganization.
2/25/2026
AI-CARE: Carbon-Aware Reporting Evaluation Metric for AI Models
Proposes AI-CARE metric incorporating carbon emissions and energy consumption alongside standard performance metrics for ML model evaluation.
2/25/2026
Intent Laundering: AI Safety Datasets Are Not What They Seem
Analyzes quality issues in AI safety datasets, finding they rely on superficial 'triggering cues' rather than genuine adversarial patterns.
2/25/2026
AI-Mediated Feedback Improves Student Revisions: A Randomized Trial with FeedbackWriter in a Large Undergraduate Course
Randomized trial showing AI-generated feedback suggestions via FeedbackWriter improve student revisions when reviewed by human TAs in economics course.
2/25/2026
Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning
Proposes symbolic alternative to GNN message-passing for more interpretable and expressive graph learning in high-stakes domains.
2/25/2026
MASPO: Unifying Gradient Utilization, Probability Mass, and Signal Reliability for Robust and Sample-Efficient LLM Reasoning
MASPO algorithm improves LLM reasoning through reinforcement learning with verifiable rewards, addressing gradient utilization and probability mass issues in existing RLVR methods.
2/25/2026
Games That Teach, Chats That Convince: Comparing Interactive and Static Formats for Persuasive Learning
User study comparing chatbots, games, and essays for persuasive learning on sustainability topics with identical content.
2/25/2026
Context-Aware Mapping of 2D Drawing Annotations to 3D CAD Features Using LLM-Assisted Reasoning for Manufacturing Automation
Uses LLM-assisted reasoning to map 2D engineering drawing annotations to 3D CAD features for manufacturing automation and process planning.
2/25/2026
How Well Can LLM Agents Simulate End-User Security and Privacy Attitudes and Behaviors?
Benchmark study (SP-ABCBench) evaluating whether LLM agents can simulate human security and privacy attitudes and behaviors for risk forecasting.
2/25/2026
Transforming Science Learning Materials in the Era of Artificial Intelligence
Examines integration of AI into science education materials, covering personalization, adaptive instruction, and accessibility in K-12 learning contexts.
2/25/2026
RPU -- A Reasoning Processing Unit
Proposes Reasoning Processing Unit (RPU) architecture to address memory bandwidth bottlenecks in LLM inference, particularly for reasoning applications with long outputs.
2/25/2026
When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators
Study revealing performance degradation when using state-of-the-art text-to-image models as synthetic training data generators for vision tasks.
2/25/2026
A Very Big Video Reasoning Suite
Benchmark suite for evaluating video reasoning capabilities in modern video models including spatiotemporal reasoning and scaling behavior.
2/25/2026
MoBiQuant: Mixture-of-Bits Quantization for Token-Adaptive Elastic LLMs
MoBiQuant enables elastic LLM deployment with token-adaptive mixture-of-bits quantization supporting dynamic precision switching at runtime.
2/25/2026
Controllable Exploration in Hybrid-Policy RLVR for Multi-Modal Reasoning
Hybrid-policy reinforcement learning framework for multi-modal LLMs with exploration strategy to prevent entropy collapse during RL training.
2/25/2026
IMOVNO+: A Regional Partitioning and Meta-Heuristic Ensemble Framework for Imbalanced Multi-Class Learning
Framework addressing class imbalance, overlap, and noise in multi-class learning using regional partitioning and meta-heuristic ensembles.
2/25/2026
Golden Layers and Where to Find Them: Improved Knowledge Editing for Large Language Models Via Layer Gradient Analysis
Layer gradient analysis method for identifying optimal layers for knowledge editing in LLMs while preserving model behavior.
2/25/2026
Model Merging in the Essential Subspace
ESM framework for merging multiple task-specific fine-tuned models using principal component analysis to reduce task interference.
2/25/2026
KnapSpec: Self-Speculative Decoding via Adaptive Layer Selection as a Knapsack Problem
KnapSpec framework reformulates self-speculative decoding as knapsack problem to optimize LLM inference throughput through adaptive layer selection.
2/25/2026
MultiModalPFN: Extending Prior-Data Fitted Networks for Multimodal Tabular Learning
Extension of TabPFN foundation model to handle multimodal tabular data integrating images, text, and tables in unified framework.
2/25/2026
Exploring Anti-Aging Literature via ConvexTopics and Large Language Models
Convex optimization-based clustering algorithm with LLM integration for analyzing biomedical literature and detecting trends in anti-aging research.
2/25/2026
The Truthfulness Spectrum Hypothesis
Study of LLM truthfulness representations across domain-general and domain-specific directions using probe generalization across five truth types.
2/25/2026
Discrete Diffusion with Sample-Efficient Estimators for Conditionals
Discrete diffusion framework using sample-efficient estimators for generative modeling over discrete state spaces with conditional probabilities.
2/25/2026
Learning to Solve Complex Problems via Dataset Decomposition
Curriculum learning approach that recursively decomposes complex datasets into simpler components using teacher-student framework with step-by-step reasoning.
2/25/2026
In-context Pre-trained Time-Series Foundation Models adapt to Unseen Tasks
Time-series foundation models augmented with in-context learning to adapt to unseen tasks without fine-tuning.
2/25/2026
QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models
QuantVLA: post-training quantization framework for Vision-Language-Action models to reduce compute and memory demands for embodied AI agents.
2/25/2026
CaDrift: A Time-dependent Causal Generator of Drifting Data Streams
CaDrift: synthetic data generator using Structural Causal Models to create data streams with controlled distributional and covariate shifts for evaluating ML methods.
2/25/2026
Emergent Manifold Separability during Reasoning in Large Language Models
Analysis of representation geometry dynamics during chain-of-thought reasoning in LLMs using manifold capacity theory.
2/25/2026
Hierarchical Molecular Representation Learning via Fragment-Based Self-Supervised Embedding Prediction
Self-supervised graph learning method incorporating fragment-level information for molecular representation learning.
2/25/2026
Momentum Guidance: Plug-and-Play Guidance for Flow Models
Plug-and-play guidance method for flow-based generative models improving sample fidelity without doubling inference cost.
2/25/2026
cc-Shapley: Measuring Multivariate Feature Importance Needs Causal Context
Method incorporating causal context into Shapley values for accurate multivariate feature importance measurement.
2/25/2026
GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training
Pre-training framework leveraging geometric data for efficient neural physics simulation with better transfer.
2/25/2026
Three Concrete Challenges and Two Hopes for the Safety of Unsupervised Elicitation
Analysis of safety challenges in unsupervised elicitation techniques for steering language models toward truthful outputs.
2/25/2026
Wasserstein Distributionally Robust Online Learning
Online learning algorithm with Wasserstein distributionally robust optimization for risk-averse sequential decisions.
2/25/2026
$\kappa$-Explorer: A Unified Framework for Active Model Estimation in MDPs
Framework for active exploration and model estimation in tabular MDPs based on coverage-based complexity.
2/25/2026
CREDIT: Certified Ownership Verification of Deep Neural Networks Against Model Extraction Attacks
Certification method for verifying DNN ownership against model extraction attacks in MLaaS systems.
2/25/2026
GauS: Differentiable Scheduling Optimization via Gaussian Reparameterization
Differentiable framework using Gaussian reparameterization for scheduling optimization in compilation and synthesis.
2/25/2026
Protein Language Models Diverge from Natural Language: Comparative Analysis and Improved Inference
Analysis of how protein language models diverge from natural language transformers with improved inference methods.
2/25/2026
Oracle-Robust Online Alignment for Large Language Models
Online alignment method for LLMs under misspecified preference feedback, extending SAIL framework.
2/25/2026
Nonparametric Teaching of Attention Learners
Attention Neural Teaching paradigm to reduce training costs for transformer-based attention learners.
2/25/2026
Elimination-compensation pruning for fully-connected neural networks
Neural network pruning technique balancing compression and information preservation in fully-connected networks.
2/25/2026
VINA: Variational Invertible Neural Architectures
Research on normalizing flows and invertible neural networks for generative modeling and inverse problems.
2/25/2026
Wireless Federated Multi-Task LLM Fine-Tuning via Sparse-and-Orthogonal LoRA
Decentralized federated learning approach for multi-task LLM fine-tuning using sparse-orthogonal LoRA over wireless connections.
2/25/2026
A Generalized Apprenticeship Learning Framework for Capturing Evolving Student Pedagogical Strategies
Apprenticeship learning framework for intelligent tutoring systems addressing sample efficiency and reward function design in educational RL.