Ioana Bica
About
I am a fourth-year PhD student at the University of Oxford and at the Alan Turing Institute, advised by Prof. Mihaela van der Schaar. My PhD research focuses on building machine learning methods for improving and understanding decision making. To achieve this, I have worked on developing causal inference methods capable of estimating the individualized effect of interventions (e.g. actions or treatments) from observational data. Causal inference enables us to perform “what if” (counterfactual) reasoning--Given the current history of observations, what would happen if we took a particular action or sequence of actions? Then, based on the estimated counterfactual outcomes, we can decide which intervention or sequence of interventions will result in the best outcome. In my work, I have built methods that can estimate such counterfactual outcomes for discrete actions taken over time, both in the presence and absence of hidden confounders, but also for continuous intervention in the static setting. Moreover, I have also been doing research at the intersection of causality, inverse reinforcement learning, and imitation learning to build machine learning methods for explaining sequential decision making on the basis of demonstrated behavior and to learn policies that are robust to distribution shifts in the deployed environment.
My research experience also includes an internship at DeepMind where I’ve been working with Jovana Mitrović on self-supervised learning and causality with the aim of learning better representations for objects in images.
Prior to my PhD, I have completed a Bachelor’s degree and a Master’s degree in Computer Science at the University of Cambridge where I have worked with Prof. Pietro Liò on multi-modal data integration and unsupervised learning for genomics data. During this time, I have also interned at Google four times.
Research interests
- Causal Inference
- Treatment Effects Estimation
- Imitation Learning
- Inverse Reinforcement Learning
- Self-Supervised Learning
- Machine Learning for Healthcare
News
Three papers accepted to NeurIPS 2021 on "Invariant Causal Imitation Learning for Generalizable Policies", "SyncTwin: Treatment Effect Estimation with Longitudinal Outcomes" and "Time-series Generation by Contrastive Imitation".
Joined DeepMind as a Research Scientist intern where I'll be working with Jovana Mitrović on self-supervised learning and causality.
Joined the Machine Learning for Healthcare Symphosium organizing committee as the Outreach Chair.
Joined the Women in Machine Learning Board of Directors.
Two papers accepted to ICLR 2021 on "Learning "What-if" Explanations for Sequential Decision-Making" and "Clairvoyance: A Pipeline Toolkit for Medical Time Series".
Our workshop on AI for Public Health was accepted to ICLR 2021.
Three papers accepted to NeurIPS 2020 on "Estimating the Effects of Continuous-valued Interventions using Generative Adversarial Networks", "Strictly Batch Imitation Learning by Energy-based Distribution Matching" and "OrganITE: Optimal transplant donor organ offering using an individual treatment effect".
Giving a talk entitled "From Longitudinal Patient Observational Data to Individualized Treatments Effects Using Causal Inference" at the Open Data Science Conference Europe. Slides here.
Joining the organizing team for the Machine Learning for Health (ML4H) workshop at NeurIPS 2020 to lead the Submission Mentorship Program.
Excited to be Teaching Assistant at the Oxford Machine Learning Summer School.
Our paper on "Ethnic and regional variations in hospital mortality from COVID-19 in Brazil: a cross-sectional observational study" was accepted in Lancet Global Health.
Our paper on the "Time Series Deconfounder: Estimating Treatment Effects over Time in the Presence of Hidden Confounders" was accepted to ICML 2020.
Organizing the Women in Machine Learning (WiML) Virtual Social at ICLR 2020.
Organizing the ELLIS agains COVID-19 event series.
Our paper on "Estimating counterfactual treatment outcomes over time through adversarially balanced representations" was selected as a spotlight presentation to ICLR 2020.