Zeyi Wen

Assistant Professor@Hong Kong University of Science and Technology (Guangzhou)·中国·OpenReview

研究方向

Efficient Large Language Model Inference · Automatic Machine Learning · Bayeisan Networks · Decision Trees · Machine Learning Systems · High-Performance Computing · Kernel Machines

6.7

TD-MoE: Tensor Decomposition for MoE Models

ICLR 2026Poster

通讯

5.0

Towards Dynamic Interleaving Optimizers

ICLR 2026Poster

二作

5.0

Representation Drift Compensation: A Zero-Cost Enhancement for LLM Decomposition

ICLR 2026Rejected

三作

4.5

Automatic Truncation Position Selection in Singular Value Decomposition for Large Language Models

ICLR 2025Withdrawn

通讯

2.5

Model-Driven Labeled Data Free Fine-tuning

合作者 (20)

Zeyi Wen

TD-MoE: Tensor Decomposition for MoE Models

Towards Dynamic Interleaving Optimizers

Representation Drift Compensation: A Zero-Cost Enhancement for LLM Decomposition

Holistic Token Efficient in Speculative Decoding: Post-use & Pre-cut

Train, Freeze or Exit: Dynamic Module-wise Fine-Tuning under Memory Constraints

Channel-Aware Mixed-Precision Quantization for Efficient Long-Context Inference

Max-Speedup Speculative Sampling: A Generic Tree Construction Principle

Efficient Fine-tuning with Decomposed Foundation Model

Interleaving Optimizers for DNN Training

Automatic Truncation Position Selection in Singular Value Decomposition for Large Language Models

Model-Driven Labeled Data Free Fine-tuning