In recent years, substantial progress has been made in maximizing model prediction accuracy. However, domains that stand to benefit most from adopting those models (e.g. healthcare, education, and policy) have been slow to adopt them because model explanation is either nonexistent or uninterpretable by non-experts. I study machine learning because I see model adoption as the most effective way to increase access to and quality of healthcare services.
Accordingly, my research interests are in developing existing discriminative and generative models (that are known to perform well in practice) for model explanation and causal inference.
Natural Language Processing
- Neural language models of entailment, modality, and causal induction
- Generative neural language models of explanation
- Counterfactual inference from observational data
- Explanation by causal description
- Variatonal Autoencoders
- Vector representation learning (e.g. of words, characters, and dependencies)
- High dimensional probability density estimation using DNNs
- Latent Variable Models