Haochen(Jason) Zhang

Perception

Scalable Web Agents LangGraph, Pytorch, AWS

Deep Learning CV pipeline for 3D Lumen and Calcification Segmentation Pytorch, CUDA

Image Processing Pipeline and Filters C++, Redis, ITK

Supervised by Prof. Mosharaf Chowdhury ▪ Investigated and experimented with asynchronous Federated Learning aggregation algorithm with buffering and staleness-aware update. Improved accuracy to 82% on OpenImage Dataset with over ~13k connected clients. ▪ Refined multi-aggregator architecture by introducing distributed Redis cache storage for client model weight update. Speeded up client selection process by searching across multiple Redis shards.

Supervised by Prof. Cyrus Omar ▪ Participated in modifying the Rust program visualization regarding variable lifetime for every program example in the official Rust book. ▪ Modified the original mdBook project for HTML generation from Markdown files to extend its functionality in generating radio element for multiple choice sections. Connected with a Actix supported back-end to create a fully functional Rust tutorial survey in investigating the effectiveness of the visualization project. ▪ Built frontend parser using Bison for customized domain-specific language to specify variable ownership transfer process.

Supervised by Prof. Dawen Cai ▪ Developed large Brainbow image file segmentation algorithm via HDF5 (Hierarchical Data Format) to enable distributed storage and decrease memory usage for file transfer and image process software. ▪ Designed MySQL Database to store segmented file metadata and FTP interface for file query and reconstruction via HDF5’s VDS (Virtual Dataset) functionality. ▪ Designed a Database Management System in C++ to store 3D Brainbow images. Programmed segmentation algorithm to transform raw image into fix-sized pages with customized header for localization. ▪ Implemented LRU cache to store frequently accessed image segments. Wrote row-sorted and column-sorted B+ tree to cache adjacent segments according to client memory usage. ▪ Designed interface for multi-user query and assemble target image. Deployed and tested database I/O performance on school server, successfully reduced visualization software rendering time by 80%.

▪ Cooperated on infrastructure research and implementation for a Citation Knowledge Graph hosted on AWS with opensourced Microsoft Academic Graph data (~100 million entries). ▪ Established multiple centrality statistics to rank paper seminality with Networkx Python library. ▪ Full Stack developed demo search engine with REST API to query and present top-ranked results from Knowledge Graph using Flask, React and MongoDB. ▪ Experimented with Hadoop’s MapReduce pipeline and GPU accelerated Graph Analysis Library (CuGraph) to achieve 100x speedup on statistics and cache file generation.

COMP 550: Algorithmic Robotics COMP 540: Statistical Machine Learning COMP 506: Graduate Compiler Construction COMP 556: Computer Networks COMP 559: Machine Learning with Graphs COMP 646: Deep Learning for Vision and Language COMP 514: Algorithm, Complexity and Approximations COMP 522: Reinforcement Learning ELEC 522: Advanced VLSI Design ELEC 553: Mobile and Embedded System

EECS 467: Autonomous Robotics EECS 485: Web Systems EECS 498: Applied Parallel Programming with GPUs EECS 491: Distributed System EECS 422: Computer Vision EECS 445: Intro to Machine Learning EECS 482: Intro to Operating System EECS 484: Database Management System EECS 281: Data structure and Algorithm