Attention Sink in Transformers: A Survey on Utilization, Interpretation, and Mitigation
Paper • 2604.10098 • Published • 84
YuJun Heo
ordidxzero
AI & ML interests
None yet
Recent Activity
Organizations
None yet
upvoted 6
papers 3 days ago
minWM: A Full-Stack Open-Source Framework for Real-Time Interactive Video World Models
Paper • 2605.30263 • Published • 62
Embodied.cpp: A Portable Inference Runtime of Embodied AI Models on Heterogeneous Robots
Paper • 2607.02501 • Published • 60
Embodied-Navigator: Point, Think, Memorize, and Align for Efficient Navigation
Paper • 2608.17512 • Published • 52
VoiceMem: Streaming Dual-Brain Memory for Real-Time Interaction
Paper • 2608.26005 • Published • 179
BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference
Paper • 2609.04971 • Published • 33
upvoted 9
papers 5 months ago
Efficient Reasoning on the Edge
Paper • 2603.16867 • Published • 22
Video Streaming Thinking: VideoLLMs Can Watch and Think Simultaneously
Paper • 2603.12262 • Published • 31
Efficient Reasoning with Balanced Thinking
Paper • 2603.12372 • Published • 152
Thinking in Uncertainty: Mitigating Hallucinations in MLRMs with Latent Entropy-Aware Decoding
Paper • 2603.13366 • Published • 95
ThinkJEPA: Empowering Latent World Models with Large Vision-Language Reasoning Model
Paper • 2603.22281 • Published • 22
LightThinker++: From Reasoning Compression to Memory Management
Paper • 2604.03679 • Published • 38
TriAttention: Efficient Long Reasoning with Trigonometric KV Compression
Paper • 2604.04921 • Published • 117
Small Vision-Language Models are Smart Compressors for Long Video Understanding
Paper • 2604.08120 • Published • 21
HY-Embodied-0.5: Embodied Foundation Models for Real-World Agents
Paper • 2604.07430 • Published • 181
upvoted a paper 6 months ago
liked a
model about 2 years ago
liked a
dataset about 2 years ago
liked 2
models about 2 years ago
