Kurt Keutzer

Full Professor@University of California Berkeley·美国·OpenReview

研究方向

Low Resource Language Translation · Natural Language Understanding at the Edge · Training Large Language Models · Efficient Deep Learning · Deep Learning

Kurt Keutzer

K-Sort Eval: Efficient Preference Evaluation for Visual Generation via Corrected VLM-as-a-Judge

SPECS: Faster Test-Time Scaling through Speculative Drafts and Dynamic Switching

ACTS: Adaptive Control for Test-time Scaling

Angles Don’t Lie: Unlocking Training‑Efficient RL Through the Model’s Own Signals

Learning Adaptive Parallel Reasoning with Language Models

Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation

Radial Attention: $\mathcal O(n \log n)$ Sparse Attention for Long Video Generation

Multipole Attention for Efficient Long Context Reasoning

Looking Backward: Streaming Video-to-Video Translation with Feature Banks

UniDrive: Towards Universal Driving Perception Across Camera Configurations

Dobi-SVD: Differentiable SVD for LLM Compression and Some New Perspectives

QuantSpec: Self-Speculative Decoding with Hierarchical Quantized KV Cache

Sparse Video-Gen: Accelerating Video Diffusion Transformers with Spatial-Temporal Sparsity

COAT: Compressing Optimizer states and Activations for Memory-Efficient FP8 Training

DrivingRecon: Large 4D Gaussian Reconstruction Model For Autonomous Driving

PixelGaussian: Generalizable 3D Gaussian Reconstruction from Arbitrary Views

SparseLoRA: Accelerating LLM Fine-Tuning with Contextual Sparsity

Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks

SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference

DrivingRecon: Large 4D Gaussian Reconstruction Model For Autonomous Driving

Interpolating Video-LLMs: Toward Longer-sequence LMMs in a Training-free Manner

VEditBench: Holistic Benchmark for Text-Guided Video Editing

SparseVLM: Visual Token Sparsification for Efficient Vision Language Models Inference