After a wonderful 4 years of PhD studies at UCSD's Artificial Intelligence Group, I have joined Carnegie Mellon University (CMU) as Assistant Professor in the Tepper School of Business, with an appointment in the Machine Learning Department (MLD).
While my research interests are eclectic, spanning both methods, applications, and social impacts of ML, there exist a few notable clusters. I am especially interested in modeling temporal dynamics and sequential structure in healthcare data, e.g., Learning to Diagnose. Additionally, I work on critical questions related to how we use ML in the wild, yielding The Mythos of Model Interpretability, and more recent work on the desirability and reconcilability of various statistical interpretations of fairness.
A grab-bag of other projects and contributions:
- Efficient exploration for RL, esp. for dialogue systems
- Algorithms for safe exploration in RL
- Establishing the precise invertibility of GANs
- Supervised disentangling the latent spaces of GANs
- Learning from noisy singly-labeled data
- Deep active learning for NLP
- Dance Dance Convolution (DNNs for DDR choreography)
I value clear, understandable scientific prose and to this end have authored / co-authored two reviews of the literature and one interactive book. Deep Learning - the Straight Dope teaches deep learning through exposition, math and code in a fully-interactive textbook written with Jupyter. The latest review explains recurrent neural networks for sequence learning tasks and can be found on the arXiv and previously, I coauthored a review on differential privacy and machine learning. Occasionally, I write for more popular media, including a feature for IEEE Spectrum magazine (February 2016). Additionally, I have been a Contributing Editor at the KDnuggets data science website. In Fall, 2016, I launched Approximately Correct, a blog aimed at bridging technical and social perspectives on machine learning. We have had some success calling out AI hype and misconceptions but the problem has only intensified.