The Computer Science Department is delighted to announce that Prof. Yu-Xiang Wang, an expert in machine learning, will be joining the faculty in Fall 2018. Here's our first interview with Yu-Xiang.
Q: How did you choose UCSB? Tell us about your story.
UCSB is a world-renown university and the computer science department that I am joining is gaining a lot of momentum, especially in AI and data science-related areas. There are some really exciting opportunities at UCSB for me to develop my career. Also, it just feels right from within! I enjoyed the campus visit, the view of a beautiful Pacific coastline, and all the technical discussions with my future colleagues. At that moment, my heart was telling me that this is where I wanted to live.
Q: What courses would you like to offer next year?
I have not completely decided on the many details of my courses, but I am likely to offer a graduate course in advanced machine learning and possibly a third or fourth year undergraduate course in convex optimization. These courses will cover both classic and modern topics in machine learning, and will be structured in ways that make it easy for students to appreciate the connections between popular techniques and how those learning algorithms work underneath.
Q: Can you tell us about your research interests?
My research interests revolve around the intersection of machine learning, statistics, and optimization. Specifically, my work focuses on developing provable and practical methods for various challenging learning regimes (e.g., high dimensional, heterogeneous, privacy-constrained, sequential, parallel and distributed) and often involves exploiting hidden structures in data (generalized sparsity, union-of-subspace, graph or network structures), balancing various resources (model complexity, statistical power and privacy budgets) as well as developing scalable optimization tools.
Q: What is your research philosophy?
I adopt the problem-driven but mathematical-oriented approach in research. In particular, I’d like to be able to formulate a real problem as concrete mathematical model, so we can get to the bottom of it (e.g., by identifying the optimal algorithm). The mathematical analysis often leads to insights about structures of the problem, which can then be used to develop faster/more reliable algorithms as well as revising the mathematical model to make it more realistic.
In general, I find my research strongly influenced by Richard Hamming’s talk “You and your research”. I highly recommend it if you haven’t seen it.
Q: What made you choose academia over industry?
We are lucky to be in an era where both academia and industry have very exciting problems to tackle. I prefer academia for the freedom to pursue something more fundamental and more rewarding in the long run. Also, I enjoy teaching and advising students.
Q: Do you have any advice for students who want to pursue machine learning studies?
I have three pieces of advice for junior PhD students in machine learning and AI:
1. Don’t just follow the trend and replicate fancy deep learning models. Go back to the basics, hone your math and keep adding to your bag of ``hammers’’, so that you will be able to smash something new and spectacular.
2. Get your hands dirty. Implement everything yourself for at least once. Keep concrete applications in mind and test your algorithm out on data sets.
3. Aim big but start small. Set up a grand long term goal and keep working towards it, but you should also keep writing down smaller milestones and share the intermediate results with the world.