Zhipeng Jia

Pretraining performance

I work on ML systems research, which improving TPU efficiency for LLMs: - Disaggregated prefill / decode serving solution - Efficient LLM serving on low-profile hardware‬ with activation sparsity - Improving Tensor Parallelism with novel distributed matmul algorithm.

https://techsysinfra.google/research/

OSA research group. Published papers in top-tier systems conferences: Telekine (NSDI 2020), Nightcore (ASPLOS 2021), and Boki (SOSP 2021).

Katana Graph is an Austin-based startup focusing on high performance graph processing and analytics, founded by UT professors Keshav Pingali and Christopher J. Rossbach.

Work with network infrastructure team.

Work with Google News team to develop the brand new machine learning-based news labeling algorithm.

I work with Technology Strategy group under the mentorship of Dr. Eric Chang and Prof. Yan Xu again. I am working on a system of automatic sleep analysis.

I work with Machine Perception team under the supervision of Dr. Hui Fang. I designed and implemented an image enhancement framework based on Google's awesome deep learning infrastructure Google Brain, which has a significant improvement over previous algorithms.

I work with Technology Strategy group under the mentorship of Dr. Eric Chang and Prof. Yan Xu. During this year, I worked on 3 projects and made major contribution in all of them: 1. Scene classification with deep neural network 2. Deep-learning-based framework for analysis of tumor pathological histograms 3. Chinese OCR (specialized for subtitles in TV programs)

Systems research, advised by Prof. Emmett Witchel. Dissertation: Designing Systems for Emerging Serverless Applications.

Yao Class (https://iiis.tsinghua.edu.cn/en/yaoclass) undergrad program