AI Research Brief
Search
Methodology
中文
Adaptive Decoding Hits 4.2x, Joint Training 10x Faster
13 selected from 376 papers
Featured
BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding
score 10
入选 HF Daily Papers; HF 热度: 71 upvotes (+4); 有代码实现; 关键词(1): lightweight
GEAR: Guided End-to-End AutoRegression for Image Synthesis
score 10
入选 HF Daily Papers; HF 热度: 30 upvotes (+4); 有代码实现; 关键词(1): text-to-image
Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs
score 12
机构: Yale; 入选 HF Daily Papers; HF 热度: 21 upvotes (+4); 有代码实现
QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents
score 9
入选 HF Daily Papers; HF 热度: 10 upvotes (+3); 有代码实现; 关键词(1): distillation
TerraDiT-$\Omega$: Unified Spatial Control for Satellite Image Synthesis with Any Geospatial Primitive
score 7
入选 HF Daily Papers; HF 热度: 4 upvotes (+1); 有代码实现; 关键词(1): synthetic data
MuSViT: A Foundation Vision Model for Sheet Music Representation
score 8
入选 HF Daily Papers; HF 热度: 5 upvotes (+2); 有代码实现; 关键词(1): fine-tuning
MemLearner: Learning to Query Context memory for Video World Models
score 7
入选 HF Daily Papers; HF 热度: 22 upvotes (+4)
DataEvolver: Self-Evolving Multi-Agent Data Construction for Text-Rich Image Generation
score 7
入选 HF Daily Papers; HF 热度: 22 upvotes (+4)
PhotoQuilt: Training-Free Arbitrary-Resolution Photomosaics via Bootstrapped Tiled Denoising
score 9
入选 HF Daily Papers; HF 热度: 24 upvotes (+4); 有代码实现
Xiaomi-GUI-0 Technical Report
score 7
入选 HF Daily Papers; HF 热度: 12 upvotes (+3); 关键词(4): deployment, fine-tuning, agentic, vision-language
Also Worth Noting
When Does Learning to Stop Help? A Cost-Aware Study of Early Exits in Reasoning Models
score 4
机构: Stanford; 关键词(2): serving, reasoning
Adaptive Cluster-First Route-Second Decomposition for Industrial-Scale Vehicle Routing
score 4
机构: Carnegie Mellon; 关键词(1): reasoning
PriorEye: Geospatial Visual Priors for End-to-End Autonomous Driving
score 4
关键词(1): real-time; 顶会接收: ECCV