Signal over volume.

Benchmarking Composed Image Retrieval for Earth Observation: From Attributes ...

Benchmarking Composed Image Retrieval for Earth Observation: From Attributes to Disaster Monitoring

Frequency-Guided Action Diffusion: Smoothing Robotic Motion via Sub-Frequency...

Frequency-Guided Action Diffusion: Smoothing Robotic Motion via Sub-Frequency Manifold Traversal

PithTrain: Redesigning MoE Training Frameworks for Agent-Task Efficiency

GPU Forecasters: Using LLMs as Selective Surrogates to Accelerate Kernel Opti...

BenHalluEval: A New Multi-Task Hallucination Benchmark for Bengali LLMs

Unlocking On-Premise LLMs for Power Systems via Boundary-Aware API Intervention

Unlocking Compositionality: How Transformers Learn to Bind Variables and Perf...

DECAT: A New Framework to Detect Spurious Correlations in Multimodal Oncology...

Unlocking Concept Binding: Why CLIP Fails and How Transformers Can Succeed

REUSERL: Boosting LLM Agent Generalization via Skill Reuse as Compression

UniAudio-Token: Bridging the Gap Between Semantic Speech and General Audio Pe...

Activation Outliers Identified as the Root Cause of Feature Death in Sparse A...

Bridging RL and Formal Verification: Value Functions as Supermartingale Certi...

Grammar-Based Symbolic Regression Discovers Thermodynamically Consistent Mate...

PARL: Learning Personalized Evaluation Rubrics from User History via Reinforc...

KAFFEE: Bridging the Gap Between Probabilistic Accuracy and Dynamical Fidelit...

Positional vs. Symbolic Attention: Why Some LLMs Fail at Long-Context Reasoning

Functional Attention: Reimagining Attention as a Mapping Between Function Spaces

Hidden Bias: VLMs Suppress Female Representations Internally Despite Neutral ...

TxFM: Surpassing Atlas-Scale Models via Curated Data and Masked Autoencoding

Unmasking the Trajectory: How Diffusion Models Generate Text from Graphs

CHARM: Giving Sensors a Voice with Multimodal JEPA for Semantic Time-Series E...

TunerDiT: Training-Free Progressive Steering for Consistent Multi-Event Video...

LongTraceRL: Enhancing Long-Context Reasoning via Trajectory-Based Distractor...