Ze M.

Software Engineer

San Jose, California, United States2 yrs 10 mos experience

Key Highlights

Developed ML models that increased ad revenue by 10%
Proposed a new framework for physics-based animation
Co-developed distributed platform for reinforcement learning

Stackforce AI infers this person is a skilled Software Engineer in AdTech and AI Research.

Contact

Skills

Core Skills

Machine LearningSoftware DevelopmentReinforcement LearningComputer Vision

Other Skills

AWS LambdaAmazon Relational Database Service (RDS)Amazon Web Services (AWS)C++Cloud ComputingCollaborative FilteringData ProcessingDeep LearningFlaskGitGo (Programming Language)JavaJavaScriptOracle DatabasePython (Programming Language)

About

A master student of Columbia University in the City of New York, major in Computer Science (software system track).My education along with experience as Software Developer Intern at EdaoQuant and SenseTime has enabled me to attain a diversified set of skills.During my professional journey, I have leveraged several skills in various facets including: -> Programming - Developed the trading platform in C++ to acquire stock data from the stock exchange.Improved user experience of backtesting with Flask and React. -> Reinforcement Learning - Designed the GAIL algorithm for virtual character animation for better performance.I have also co-developed a distribtued platform for reinforcement learning at SenseTime. -> Computer Vision - Helped building HAKE, a large-scale human action dataset and designed and tested the state-of-the-art algorithms on the dataset.

Experience

Tiktok

2 roles

Software Engineer

Jan 2023 – Dec 2024 · 1 yr 11 mos · San Jose, California, United States

TikTok, E-commerce Shop Ads
Ads Ranking & Infra
Designed, built and iterated an ML model with training and serving pipelines from scratch to predict user purchasing amounts in Ads ROAS optimization. Increased value-based ad revenue by 5% in A/B testing
Developed and iterated on several Collaborative Filtering retrieval methods such as Item-CF, and real-time signal-based retrieval such as precise retargeting, resulting in a 5% increase in shop ad revenue in A/B testing.
Led efforts to refactor infrastructure with the organic recommendation team to establish a unified nearline retrieval system with shared user signals, retrieval methods, and interaction history. This refactoring led to a series of launches that boosted ad revenue by 10%.
Bidding & Auction
Initiated, designed, and developed a pacing strategy to balance costs between GSP and GFP placements with the Search Ads team, resulting in an 18% revenue increase and a 4% gain in ROAS.
Investigated, identified, and resolved a series of issues in advertiser budget utilization through optimizations in the bid exploration, real-time cost calculation, and automated bid capping adjustments.
AIGC
Researched, prototyped, and open-sourced a series of work (https://magic-me-webpage.github.io) for diffusion-based algorithms to preserve identity in AIGC videos.

C++Deep LearningMachine LearningSoftware Development

Software Engineer Intern

May 2022 – Nov 2022 · 6 mos · New York, United States

Shopping Ads
Develop the bidding infra and algorithm to increase the revenue.
Design the machine learning model on PCTR.
Iterate over the UI/UX features in the frontend.

Python (Programming Language)JavaSoftware DevelopmentMachine Learning

Berkeley artificial intelligence research

Research Assistant

Jan 2020 – Dec 2020 · 11 mos · Berkeley, California, United States

Proposed and implemented a new adversarial imitation learning framework for physics-based animation.
AMP: Adversarial Motion Priors for Stylized Physics-Based Character Control
Xue Bin Peng*, Ze Ma*, Pieter Abbeel, Sergey Levine, Angjoo Kanazawa (*Joint first authors.)
ACM Transactions on Graphics (Proc. SIGGRAPH 2021)

TensorflowPytorchReinforcement LearningComputer Vision