Language Technologies Institute Colloquium


While deep learning produces supervised models with unprecedented predictive performance on many tasks, under typical training procedures, advantages over classical methods emerge only with large datasets. The extreme data-dependence of reinforcement learners may be even more problematic. Millions of experiences sampled from video-games come cheaply, but human-interacting systems can’t afford to waste so much labor. In this talk I will discuss a series of works, each aimed at increasing the efficiency of learning from human interactions. Specifically, I will discuss efficient dialogue policy learning, deep active learning for named entity recognition, learning from noisy singly-labeled data, and active learning with partial feedback.

After graduating from UCSD, Zachary recently joined Carnegie Mellon as an 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 concerning how we use ML in the wild, recently yielding The Mythos of Model Interpretability, and new work on the desirability and reconcilability of various statistical concepts related to fairness. Other projects include generative modeling, active learning, reinforcement learning, crowdsourcing, and fun applications of deep learning to music, video games, and beer review generation. I value clear, understandable scientific prose and to this end have authored / co-authored two reviews of the literature (on RNNs and differential privacy) and one interactive book (in progress), teaching deep learning through exposition, math, and code using Jupyter notebooks.  In late 2016, I launched Approximately Correct, a blog aimed at bridging technical and social perspectives on machine learning.

Instructor: Graham Neubig

For More Information, Please Contact: