Bobby

Education

• PhD candidate, Computer Science.
• Advisor: Prof. Fredrik Kjolstad.

• B.S. in Electrical Engineering and Computer Science (EECS).

• International Baccalaureate Diploma. Advanced Science Program.

Experience

• Worked on understanding how data quality affects fairness. Developed and implemented an approach to reduce fairness violations in multi-class classification via dataset reweighting.
• Methods and results submitted and accepted at DataPerf @ ICML 2022.

• Building systems for machine learning—data management, model debugging, experiment reproducibility, and model explainability with Rolando Garcia, advised by Prof. Joe Hellerstein and Prof. Joseph Gonzalez.

• Head TA for CS 162: Operating Systems, TA for CS 188: Artificial Intelligence.
• Taught weekly discussion sections of ~35 students, held weekly office hours, homework and project parties.
• Led exam reviews. Wrote and graded exams for upper-division classes.
• Contributed and reviewed course contents, including lecture notes, homework, projects, and exams.
• Improved course infrastructures and revamped the course website for CS 188.
• Average teaching rating of 6.32/7.00, 0.55 above the department average.

• Research in deep reinforcement learning, working on safe RL, efficient policy optimization with safety constraints with Ashwin Balakrishna and Brijen Thananjeyan, advised by Prof. Ken Goldberg.
• Developed methods to improve the scalability and stability of model-based RL algorithms based on model predictive control

• Applied domain randomization techniques to train robust domain-invariant object recognition and robot grasping models for surface declustering tasks, by increasing simulated variability to improve the generalizability to the real-world.

• Answered conceptual questions, guided students through problems and debugging during weekly office hours, labs, and project parties.

Publications

Honors & Awards

Projects

• A record-replay system for efficient hindsight logging in model training.

• Class website for CS 188: Artificial Intelligence.

• A deep RL agent that plays Gomoku (a.k.a. Five in a Row) through random and self plays.

• Placed in the top 2% of 300+ teams in NP-Hard constrained graph partitioning contest using a combination of METIS multilevel k-way partitioning scheme, greedy vertex swaps, local search, and simulated annealing.

• Built a voice-controlled electric car and implemented closed-loop control system to reduce disturbance and error in car movements. Employed signal filtering, PCA and K-means clustering to distinguish voice commands and determine the spoken words.

• A fast and versatile QR Code Generator.

• An addictive and beautiful-looking (gorgeous on the iPhone X OLED display) arcade game with tilt-free gameplay, a nice BGM, and cool sound effects.

• A language-learning-oriented social networking app built using Swift 3 and Firebase.

• Over 25,000 views per month.
• Compilation of notes and guides for high school students to study and prepare for IB exams.

• A practical and highly effective app for learning new Spanish words, with over 15k downloads and more than 1,000 active users.

• Official website for TEDx events organized at ACS (Independent), Singapore.