ML & AI for Streaming Discovery VenturaOS @ The Trade Desk | Ex-Haystack News, Microsoft, Motorola Solutions
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
Yi has been a passionate ML practitioner/applied scientist for 10 years. He deeply believes in a data-driven approach and are excited about applying ML/AI to solve the world's problems. He has a solid understanding of ML problems including classification, regression, ranking and have experience on a wide range of topics such as Deep Learning, Reinforcement Learning, Recommender System, Natural Language Processing and display ads optimization.
University at Buffalo
Ph.D., Computer Science
January 1, 2006 – January 1, 2012
Beijing University of Posts and Telecommunications
B.E., Computer Science
January 1, 2002 – January 1, 2006
University of Electronic Science and Technology of China
Part time self-study, Computer Science & Engineering, Electronics
January 1, 1993 – January 1, 1995
Sichuan Vocational College of Chemical Technology
Diploma, Chemical Analysis (4-year program for technical diploma)
January 1, 1991 – January 1, 1995
The Trade Desk
Senior Staff Data Scientist
July 1, 2025 – Present
Bellevue, Washington, United States · Hybrid
Haystack News
Staff Machine Learning Engineer
August 1, 2023 – December 1, 2024
Redmond, Washington, United States · Remote
Microsoft
Sr Machine Learning Scientist, Tech Lead
April 1, 2020 – November 1, 2023
Microsoft
Sr Machine Learning Scientist
March 1, 2017 – March 1, 2020
Microsoft
Machine Learning Scientist II
December 1, 2014 – March 1, 2017
Microsoft
Software Development Engineer II
April 1, 2013 – December 1, 2014
Motorola Solutions
Senior Staff Research Engineer
November 1, 2012 – March 1, 2013
Schaumburg, Illinois
University at Buffalo
Research Assistant
June 1, 2009 – August 1, 2012
On-site
University at Buffalo
Research Assistant
June 1, 2008 – October 1, 2008
On-site
University at Buffalo
Teaching Assistant
August 1, 2006 – May 1, 2011
On-site
Yunnan Yuntianhua Co., Ltd.
Chemical Analyst
August 1, 1995 – August 1, 2002
Zhaotong, Yunnan, China · On-site
Implementation of Feature Truthing Tool for Collecting Handwriting Data
November 1, 2011 – December 1, 2011
Goals: 1). To aid document examiners to provide ground truth for handwriting samples; 2). To differentiate between hand-print and cursive writing styles; 3). To allow users to load any images forward/backward; 4). To allow users to verify and update the truthed results; 5). useful tool for getting training data for handwriting identification. Platform: Visual Studio Techniques: C++, win32, MFC.
Structure Learning of Bayesian Networks
November 1, 2011 – March 1, 2012
1). Proposed O(n^2 log(n)) algorithms for learning the structure of Bayesian networks; 2). Experiments showed the algorithm is more accurate and efficient than optimized branch and bound algorithm (ICML09, http://dl.acm.org/citation.cfm?id=1553374.1553389). Techniques: Chi-squared tests, greedy search, KL divergence.
Statistical Models of Computing Likelihood Ratio for Identification
May 1, 2011 – August 1, 2011
1). Proposed a conjecture that the likelihood ratio (LR) may be decomposed as the product of a distance factor and a rarity factor; 2). Evaluated the conjecture with several data types (continuous, binary, multi-nomial and graphs) and modalities (handwriting, footwear evidence); 3). The proposed framework is accurate, efficient and generally applicable (more accurate than distance method and joint distribution method). Techniques: hypothesis testing, density estimation, clustering, learning Bayes nets
Shoeprint Clustering, Retrieval and Identification
June 1, 2009 – August 1, 2010
Goals: 1). To find similar shoeprints from database for a given query shoeprint; 2). To determine whether or not a crime scene footwear mark and a known print came from the same source. I designed algorithms to extract robust features, constructed Attributed Relational Graph (ARG) to represent each shoeprint, proposed a distance measure between ARGs based on linear programming techniques. Our shoeprint image retrieval system outperformed state-of-the-art image matching algorithms such as SIFT, Shape Context and other existing footwear retrieval systems. Clustering based on recurring patterns speeds up the retrieval 10+ times. A prototype was demonstrated during the presentation at Foster & Freeman Ltd., UK. Techniques: randomized algorithms, graph matching and graph distance measure, CBIR, linear programming (optimization), sensitivity analysis.
Information Fusion Architecture for Safety Applications in Vehicular Networks
April 1, 2009 – June 1, 2009
1). Proposed a novel architecture for information fusion in vehicular networks (VANET); 2). Devised a multi-level fusion algorithm that performs info fusion/aggregation at message level, hazard level, and decision level; 3). Designed test scenarios to evaluate performance using simulator STISIM Drive.
Driver Mobility Profiling in Vehicular Networks
June 1, 2008 – September 1, 2008
1). Collected synthetic vehicle trace by (probabilistic) modelling driver visitation pattern, programming with Google Maps APIs and parsing response messages to extract way-points; 2). Learned Hidden Markov Model (HMM) to profile driver mobility; 3). Predicted driver locations and evaluated performance with synthetic/real traces. Techniques: JavaScript, HMM, Expectation–maximization, Clustering, Mixture Models.
Fall 2020 MLADS Conference Volunteer
Microsoft
June 24, 2026 – Present
Data Analytics: Dashboards vs. Data Stories
June 24, 2026 – Present
Cultural Fit Analysis
The candidate's career progression from Research Assistant to Senior Staff Data Scientist, with significant tenure at Microsoft, indicates stability and a commitment to growth within an organization. The diverse range of projects, from academic research in computer vision and Bayesian networks to industry applications in recommendation systems and ad monetization, suggests adaptability and a broad interest in various ML domains. The focus on impactful, user-facing systems aligns with a results-oriented culture. The transition from chemical analysis to computer science demonstrates a strong drive for continuous learning and career evolution.
Soft Skills & Operational Fit
The candidate's extensive experience as a Senior Staff Data Scientist, Staff Machine Learning Engineer, and Sr Machine Learning Scientist/Tech Lead at major companies like Microsoft and The Trade Desk suggests strong leadership, collaboration, and problem-solving skills. The descriptions of improving content recommendation services and leading teams imply operational effectiveness and the ability to drive projects to completion. The academic background and research projects also point to strong analytical and independent problem-solving capabilities.