Signal over volume.

CRAM: Balancing Format Isolation and Visual Reuse in Multimodal Continual Ins...

CRAM: Balancing Format Isolation and Visual Reuse in Multimodal Continual Instruction Tuning

SafeSteer: Minimizing Alignment Tax via Localized On-Policy Distillation

RAID: A No-Regret Framework for Adaptive Incentive Design in Nonlinear Games

SkillHarm: Uncovering Critical Vulnerabilities in the Lifecycle of Agentic Sk...

IntraShuffler: Defending Heterogeneous DP Federated Learning Against Gradient...

AdaCodec: Redefining Video MLLM Interfaces via Predictive Visual Coding

CLINENV: A New Benchmark for Evaluating LLMs as Longitudinal Inpatient Physic...

Perception-Judge: Mitigating Visual-Textual Conflict in Multimodal LLM Evalua...

PROTOADA: Solving Format-Induced Forgetting in Multimodal Continual Instructi...

Langevin Speculative Dynamics: Accelerating Molecular Dynamics via Parallel V...

Verifiable Belief-Space Safety Filters: Reducing Conservativeness in Interact...

Speculative Pipeline Decoding: Achieving Zero-Bubble Acceleration via Pipelin...

T5: Linearizing Vision Transformers via Test-Time Training for Efficient High...

Brain-IT-VQA: Decoding Visual Questions Directly from Human Brain Activity

ESPO: Reducing LLM Reasoning Compute by Early-Stopping Failing Trajectories

CRAFTER: A Multi-Agent Harness for Editable Scientific Figure Generation and ...

Harness Updating is Not Harness Benefit: Disentangling Evolution in LLM Agents

MetaPlaq: A Multimodal Atlas Mapping Cellular Drivers of Human Atherosclerosis

SurGe: Enhancing Monocular 3D Reconstruction via Neighborhood Attention and G...

One-Forcing: Achieving High-Fidelity One-Step Autoregressive Video Generation

DEMON: Turning Diffusion Models into Real-Time Playable Musical Instruments

GCPO: Enabling Fine-Grained Token Credit Assignment in Discrete Policy Optimi...

VisualThink-VLA: Achieving High-Accuracy, Low-Latency Robot Control via Visua...

The Physical AI Inference Gap: Why Faster GPUs Don't Always Mean Faster LLM D...