Director at Facebook AI
Strong Technical Leader with R&D experience/background on multi-disciplinary technical skills as: Machine Learning, Deep Learning, Personalization, Knowledge Graph, Embedding, Fraud, Risk Management, User Profiling, User Behavior Modeling, Business Intelligence, Optimization, Statistics, Computational Intelligence, with very large scaled data sets.
Talks and Events
2022 Talk: Forefront Of Research And Pragmatic Production Enablement
The AI Applied Research mission is to differentiate the step function of new bets that will bend the curve, in order to strongly impact the long-term direction of products.
Recommendation products, as well as monetization, are evolving to be more focused on interest- driven distribution, user-generated content, and ecosystem building. Each of the products face different challenges. For Applied research, we possess the capability to deeply understand people’s problems, and summarize or translate them to the proper technical investment in order to solve the problem holistically.
Considering the very complicated ecosystem and different product focus, our Applied Research involves multiple areas that are essential to our daily problem solving. These researchers enable AI to become the engine of tech innovation to the industrial revolution. I would like to touch on a few of them, such as Content Understanding, User Understanding, Training Scalability and Realtimeness, which are deeply involved in the extensive research about Graph Learning, Reinforcement Learning, Model Architecture and High Performance Computing.
During the research stage, more attention is paid on how to improve modeling performance through algorithm evolution, inventing new approaches and the fusion of different ideas. When reaching the enablement stage, we are facing another layer of challenges. Product enablement is critical to applied research since our ultimate goal is to significantly impact the projects with these tech innovations. It is not the last mile, but could be half way. In general, when holistically solving industrial problems, we have to think about the enablement challenges in the frontload mode, sometimes even at the early research stage. Designing the system towards flexibility allows us to upgrade components in response to the changing technical landscape and business needs, which are also essential to our long term applied research.