profile_picture
Alan Zhu
CS Student, Carnegie Mellon University

I am a senior undergraduate at Carnegie Mellon University (CMU), where I major in Computer Science with an additional major in Mathematical Statistics. I currently work in the Catalyst Lab at CMU under the supervision of Prof. Zhihao Jia. Previously, I have worked in the Interactive Structures Lab at CMU under the supervision of Prof. Alexandra Ion. I have also worked on research under the supervision of Prof. Qiaozhu Mei at the University of Michigan.

My research interests are in the field of Machine Learning (ML) generally, with a focus on improving usability through efficiency, explainability, and accessiblity, and in Natural Language Processing.

Interests

  • Machine Learning
  • Efficient ML
  • Systems for ML
  • Natural Language Processing

Academics

Carnegie Mellon University
2020 - 2024 (expected)
B.Sc. Computer Science, additional major in Mathematical Statistics
4.0/4.0 GPA | Dean's List, High Honors (Fall 2020 - Present)

Manuscripts

Accelerating Retrieval-augmented Language Model Serving with Speculation, 2024, Submitted to ICML
Zhihao Zhang , Alan Zhu , Lijie Yang , Yihua Xu , Lanting Li , Phitchaya Mangpo Phothilimthana , Zhihao Jia
arxiv
Examining Shortest-Path Distances to Samples in Graphs, 2024, Submitted to ICWSM
Alan Zhu , Jiaqi Ma , Qiaozhu Mei
pdf
SpecInfer: Accelerating Generative Large Language Model Serving with Speculative Inference and Token Tree Verification, 2024, To appear at ASPLOS 2024
Xupeng Miao , Gabriele Oliaro , Zhihao Zhang , Xinhao Cheng , Zeyu Wang , Rae Ying Yee Wong , Alan Zhu , Lijie Yang , Xiaoxiang Shi , Chunan Shi , Zhuoming Chen , Daiyaan Arfeen , Reyna Abhyankar , Zhihao Jia
arxiv
Robotic Metamaterials, 2024, To appear at CHI 2024
Zhitong Cui , Shuhong Wang , Violet Yinuo Han , Tucker Rae-Grant , Willa Yunqi Yang , Alan Zhu , Scott E Hudson , Alexandra Ion

Research Projects

See CV for more details.
Accelerating Retrieval-augmented Language Model Serving with Speculation
PI: Prof. Zhihao Jia. We propose a method to improve the latency of Retrieval-augmented Language Model systems through adaptive speculation.
Examining Shortest-Path Distances to Samples in Graphs
PI: Prof. Qiaozhu Mei. We propose a framework to predict the distribution of shortest-path distances to a sample of nodes in a graph, generalizable to a variety of sampling methods. The framework has applications in sample selection for training graph neural networks and other studies.
SpecInfer: Accelerating Generative Large Language Model Serving with Speculative Inference and Token Tree Verification
PI: Prof. Zhihao Jia. We propose a system to improve LLM serving latency by speculatively decoding tokens. I provided a proof demonstrating our decoding method is uniformly superior to a naive implementation.
Robotic Metamaterials
PI: Prof. Alexandra Ion. Robotic metamaterials are grids of deformable or rigid square cells which can produce motions by actuating some cells able to control their own deformation. I built an algorithm designing robotic metamaterials given desired mechanism motions.

Teaching

See CV for responsibility details
TA for 15251 Great Ideas in Theoretical Computer Science
Head TA: Fall 2023 - Present
Associate Head TA: Spring 2023
TA: Fall 2021 - Fall 2022
public course website (spring 23)

Miscellaneous

In high school (at Greenhills School) I was active in math and robotics competitions. I qualified for the USAJMO in 2016 and the USAMO in 2019. I also served as programming team lead for the robotics team (FRC Team #5530 - The Greenhills Lawnmowers), which qualified for the FIRST World Championships in 2019.