Bokun Wang

Bokun Wang

Postdoctoral Researcher at UT Austin

I am a Postdoctoral Fellow in the Department of Electrical and Computer Engineering (ECE) at The University of Texas at Austin, working with Prof. Diana Marculescu and the EnyAC group. Before that, I received my Ph.D. degree in Computer Science at Texas A&M University, advised by Prof. Tianbao Yang.

My research interest lies in Hardware-efficient Generative AI.

Email: bokun.wang@utexas.edu

Research

Hardware-efficient GenAI Inference

Improving hardware efficiency (latency, memory, energy consumption, etc.) of video and language GenAI inference while preserving generation fidelity.

Data/Memory-efficient Contrastive Learning

Memory-efficient and data-efficient (multi-modal) contrastive learning with convergence and generalization guarantees.

Efficient Imbalanced and Multi-instance Medical Image Analysis

Memory-efficient provable algorithms for high-res, imbalanced multi-instance medical image analysis.

Communication-efficient Distributed and Federated Learning

Communication-efficient and provably convergent algorithms for distributed and federated learning.

Recent Work (Full List)

Teaser for Publication 1

DARE: Diffusion Language Model Activation Reuse for Efficient Inference

Natalia Frumkin, Bokun Wang, Hung-Yueh Chiang, Chi-Chih Chang, Mohamed S. Abdelfattah, and Diana Marculescu

ArXiv Preprint, 2026

Training-free Latent Inter-frame Pruning with Attention Recovery

Dennis Menn, Yuedong Yang, Bokun Wang, Xiwen Wei, Mustafa Munir, Feng Liang, Radu Marculescu, Chenfeng Xu, and Diana Marculescu

To appear in Conference on Neural Information Processing Systems (NeurIPS), 2026

Teaser for Publication 3

A Geometry-Aware Efficient Algorithm for Compositional Entropic Risk Minimization

Xiyuan Wei, Linli Zhou, Bokun Wang, Chih-Jen Lin, and Tianbao Yang

In the Forty-third International Conference on Machine Learning (ICML), 2026

Teaser for Publication 4

ELANA: A Simple Energy and Latency Analyzer for LLMs

Hung-Yueh Chiang, Bokun Wang, and Diana Marculescu

Technical Report, 2025

Awards and Recognitions

Teaching