Tianyu Zhang 张天宇

Tianyu Zhang 张天宇

Undergraduate AI Major. Dreamer. Doer.

Carnegie Mellon University

ByteDance IES


I am a junior at Carnegie Mellon University School of Computer Science. I major in Artificial Intelligence and minor in Computational Finance.

My current interest lies at the intersection of machine learning and scalable, distributed systems. I interned at the recommender systems team at ByteDance summer 2020 and devised ML algorithms and systems for billion-DAU products.

I’m actively seeking internship opportunities for Summer 2021 during which I wish I could explore new knowledge and different career opportunities. Specifically, I’m looking for software/algorithm engineering or quantitative analysis roles. Please shoot me an email or direct message if you have such opportunities!


  • Scalable/Distributed ML Systems
  • Recommender Systems
  • Person/Object Re-identification
  • Learning Theory
  • Interpretable ML


  • B.S. in Artificial Intelligence, Expected 2022

    Carnegie Mellon University

  • High School Diploma, 2018

    Beijing National Day School & Wasatch Academy

Languages & Tools

with self-evaluated levels of expertise

























Machine Learning Algorithm Intern


Jun 2020 – Aug 2020 Beijing

Live Streams Recommendation, Graph Embedding:

∗ Improved the performance of the internal ML trainer (C++) by reducing communication overhead. Reduced the mini-batch forward latency by over 100% in certain training circumstances that involve graph embedding.

∗ Extracted relations from Petabyte log data to build distributed user-author graphs using MapReduce and designed graph encoders to train an embedding of graph nodes using Tensorflow and ByteDance ML API.

∗ Integrated end-to-end embedding to CTR prediction and significantly increased online user staytime (+3.5%), comment rate (+3.8%), and other metrics in AB tests for Douyin and TikTok.

Systems for Engineering Efficiency:

∗ Designed and developed a system from scratch in Django with RESTful APIs that creates and manages alerts for 100+ online models of 5 products and presents model health status on a dashboard. Attracted 70+ internal users and developers. Reduced the usual day-long response time to less than an hour in a recent dataflow accident.

∗ Developed a pipeline that analyzes the importance of 300+ features in a Monte Carlo fashion and performs feature modification on Terabyte model checkpoints on clusters based on the results. Saved 2hr+ manual labor per iteration and 35k+ core-hour computing resources in total than hand-tuning.


Machine Learning Intern


May 2019 – Jul 2019 Beijing

Baggage Re-Identification for Smart Security Inspection: Extensively investigated past architectures for person and vehicle re-identification (re-ID) tasks. Implemented multiple architectures and evaluated their performance for baggage re-ID. Achieved a 0.76 accuracy of CMC rank-1 (improving the baseline by a 0.34 margin in accuracy and about 100x in speed) on the overall re-ID task (image retrieval from the 500-baggage gallery) on the Multi-View Baggage dataset.

Testing Framework Development: Developed a fork upon the open-source deep-person-reid framework for the team’s future research and deployment workflow, including features such as activation visualization, training checkpoint management, Comet.ml integration, and CLI tools.


Head Academic Staff

TechX Academy

May 2018 – Aug 2019 Shanghai

Instructor of Theory of Computation: Co-taught the 5-day seminar on theoretical computer science, covering topics such as discrete math fundamentals, computability (DFA and TM), and efficiency (polynomial reduction and P vs. NP). Designed slides, handouts, and assignments.

TA for AI Courses: TA-ed 10-day main courses. Designed homework and assessments for neural net basics, Markov Decision Process, Q-learning in Deep Reinforcement Learning (2018), and feature extraction, SVM, DP & Seam Carving in Computer Vision (2019). Led Q&A sessions and graded homework.

Academy Experience Design: Directed the Academic Team of TechX 2019. Led the processes of course design, staff search, and infrastructure setup for 5 courses.




Apr 2018 – Apr 2019 Beijing

Product Management: Assisted the development of the platform from scratch using TypeScript and Node.js; generated user experience reports and feature addition proposals; beta-tested the platform with 100 users.

Business Operations: Drafted business plans and pitched to Tsinghua alumni investors, ZhenFund, and various other VCs; Invited 15+ Chinese corporations and organizations to enter Fintern.

Relevant Courses

Courses I’ve taken over the years. Items marked * are on-going.

Computer Science

  • Algorithms Design and Analysis*
  • Parallel and Sequential Data Structures and Algorithms
  • Introduction to Computer Systems
  • Great Ideas in Theoretical Computer Science
  • Functional Programming
  • Principles of Software Construction
  • Competitive Programming

Artificial Intelligence

  • Scalable Machine Learning*
  • Computer Vision*
  • Introduction to Machine Learning
  • AI: Representation and Problem Solving


  • Multivariate Calculus
  • Probability Theory
  • Linear Algebra
  • Discrete Math
  • Modern Regression*
  • Introduction to Mathematical Finance


More to come…

Time Series Analysis Framework

This framework is aimed to provide analytical tools and operations on time series from any domain and allows plugins to be written to import time series from custom sources and visualize them in custom ways. Written in Java with GUI constructed with Java Swing.

Malloc (Dynamic Memory Allocator)

Pinnacle of Intro to Comp Sys. A dynamic memory allocator with segregated list implementation. Supports 64-bit address space and achieves 74.6% average utilization.

Carcassonne (Board Game)

A multi-player tile-laying board game. GUI constructed with Java Swing.

Small Assembly Interpreter

Parses and simulates execution of Small Assembly (a 26-instruction subset of the standard Assembly language) on a virtual machine. Written in Java with an extensible design.

deep-person-reid MVB Fork

Deep learning re-identification in PyTorch. A forked version with additional functionalities and MVB interface.

MVB ReID with Verification Scheme

Models and experiments for the baggage re-id task using the verification scheme.


K-mer search enabled fast bacterial DNA sequence retrieval with DNA sequence analysis of Pittsburgh rivers.


A group-based crowdsourcing notification system. Never miss out on an impromptu event again! Built at CMU TartanHacks 2019.


A notepad built from Python Flask.

City Selection for Int’l Conferences

Venue Selection of International Conferences Based on Time Difference and Flight Fatigue: Explored the influence of jet lag and flight fatigue on travelers; built models to provide strategies for international conference venue selection. (IMMC2017, Finalist, MATLAB/Mathematica)

Language Distributions Prediction

Predicting the Distributions of Language Speakers and Its Application in Office Site Selections: Borrowing from the SIR model from epidemiology and the gravitational model from physics, built a model to analyze the spread of languages; utilized AHP to analyze and offer recommendations of global office setup for international corporations. (MCM2018, Meritorious, Python/MATLAB)

Recent & Upcoming Talks

The Theory of Computation

Understand the theoretical underpinnings of computer science.
The Theory of Computation

Not a Typical Talk on Recommender Systems

Understand recommender systems from a less technical perspective
Not a Typical Talk on Recommender Systems


Let’s talk about anything :)