Bay Area Tech Economics Seminar Series: Maria Dimakopoulou
This event is open to:
Thursday, December 7th, the Bay Area Tech Economics Seminar Series will feature Maria Dimakopoulou, Director of ML Engineering and Head of Homepage Personalization at Spotify, and a Stanford alum. Her talk is titled: Causal Adaptive Learning for Recommendations.
Doors will open at 6PM, a reception with light refreshments will follow the talk. Location details for each session will be provided to registered participants. Sign up here to be included on the mailing list.
Sequential decision making and accurate model estimation from adaptively collected data lie at the heart of personalized recommendations. If we want to have reliable decision making in practical recommender systems which adapt to users’ feedback via contextual bandit or reinforcement learning algorithms, we have to get model estimation right. In this talk, we will cover how to incorporate state-of-the-art methods from the causal inference literature in the model estimation of recommender systems and how to pair them with efficient exploration strategies such as Thompson Sampling. Further, we will discuss the personalization performance gains that this approach unlocks in the presence of real-world challenges such as selection bias, covariate shift, model-misspecification and bias due to adaptive data collection.
Maria Dimakopoulou is Director of ML Engineering and Head of Homepage Personalization at Spotify. The Homepage Personalization organization consists of 60+ Engineers & Engineering Managers, spanning ML Engineering, ML Research, Backend, Data & Web and is responsible for generating, ranking and distributing personalized recommendations across music, podcasts and audiobooks on the Spotify Homepage of the 574+ million listeners. The teams within the Homepage Personalization organization innovate on a range of ML disciplines – contextual bandits, reinforcement learning, attention-based deep networks, multi-objective optimization and causal inference – as well as ML infrastructure, data and systems at scale, with the goal to serve recommendations that maximize the listeners’ joy, while supporting Spotify’s strategic & business needs. Before joining Spotify, she was at Netflix, first driving causal recommendation initiatives and subsequently building and leading the Adaptive Experimentation team, which was a cross-functional team of researchers, engineers and data scientists focused on delivering new experimentation capabilities at Netflix (adaptive bandit-based online tests instead of A/B tests; valid inference from adaptive online tests; ML-based short-term proxy metrics of long-term online metrics etc.). Prior to Netflix, I did my PhD on Reinforcement Learning and Causal Inference at Stanford, where I was advised by Benjamin Van Roy and Susan Athey. Before her PhD at Stanford, she worked at Google Research on the design and deployment of large-scale optimization algorithms for Google Technical Infrastructure and Google Ad Exchange.