Google DeepMind - Image Understanding

Google Research - Perception - Image Understanding

* Analyzed, designed, modeled, and productionized the digital audience measurement product including demographics, watch time, unique viewers, and topics for over 1B social videos in the world using machine learning* Unique audience estimation via probabilistic counting* Social video category classification and topic assignment with our in-house content graph

Researched in the fields of theoretical particle physics, cosmology, and astrophysics. Built mathematical models to explain the phenomena in the universe such as the matter-antimatter asymmetry and dark matter. Tested theoretical models against observational data like Planck experiment via numerical calculations. Published and presented our results in peer-reviewed journals and conferences.

Achieved the Kaggle Competitions Master with the highest rank 400 of 81625 within a year of join on Kaggle. Won 1 gold and 3 sliver medals in 4 competitions. 1. “Web Traffic Time Series Forecasting” Gold Medal (Top 1%, 11th of 1095) 09/2017- Designed 1D convolutional neural network (CNN) to predict future traffic to 145k Wikipedia pages- Developed the idea of group model and applied ensemble learning to lower the error by 50%2. “Recruit Restaurant Visitor Forecasting” Silver Medal (Top 1%, 17th of 2158) 02/2018- Trained multiple LightGBM models to predict future visitors (time series) to the restaurants in Japan- Engineered features from existing data to improve performance of the models3. “Porto Seguro’s Safe Driver Prediction” (binary) Silver Medal (Top 4%, 164th of 5169) 11/2017- Trained multiple XGBoost and embedding neural network models for predictions- Collaborated with teammates to fulfill the project efficiently within the time limit4. “Santa Gift Matching Challenge” (optimization problem) Silver Medal (Top 6%, 23th of 428) 01/2018- Combined the min-cost max flow problem and the Monte Carlo (MC) swapping algorithm to solve the constrained optimization problem. - Worked in C++ and Google Ortools

Worked in the fields of theoretical particle physics, cosmology, and astrophysics. Explored possibility to explain the matter-antimatter asymmetry of the universe using the Higgs field relaxation in the early universe. Built computer programs to solve the system of renormalization group equations of the Higgs couplings numerically. Analyzed the time-evolution of the Higgs field both analytically and numerically. Resulted in 6 journal publications with 1 in Physical Review Letters.

Led 15 discussion and review sections for quantum mechanics, mathematical method in physics, and other physics classes with class size about 50. Given few substitution lectures on quantum field theory.

Explored observational constraints on f(R) modified gravity theory in cosmology. Computed and predicted the spectrum of the gravitational waves generated by the f(R) models. Resulted in 2 journal publications.