Human-in-the-loop Probabilistic Graphical Modeling

Abstract: Generative models are venerated as full probabilistic models that randomly generate observable data given a set of latent variables that cannot be directly observed. They can be used to simulate values for variables in the model, allowing analysis by synthesis or model criticism, towards an iterative cycle of model specification, estimation, and critique. However, many datasets represent a combination of several viewpoints -- different ways of looking at the same data that leads to various generalizations. For example, a corpus that has data generated by multiple people may be mixtures of several perspectives and can be viewed with different opinions by others. It isn't always possible to represent the viewpoints by clean separation, in advance, of examples representing each perspective and train a separate model for each point of view. In this talk, I introduce a mixed-initiative technique to (1) extract mappings between machine-learned representations and perspectives of human experts, and (2) estimate graphical models that afford multiple views of the same dataset. I will explore human-in-the-loop estimation of latent variable models in their configuration, parameter and evidential spaces. I will discuss how this paradigm can be applied in three health applications, namely imbuing the perspectives of experts into latent variable models that analyze adolescent distress, probe atypical angina in cardiology, model crisis counseling, and investigate self-harm in clinical psychology.

Bio: Dr. Karthik Dinakar is a computer scientist specializing in machine learning, natural language processing, and human-computer interaction. His doctoral thesis involved representing human knowledge formally within probabilistic graphical models as a way of making algorithms learn from both the data as well as human expertise within their inference training loops. Karthik was a Reid Hoffman Fellow at Massachusetts Institute of Technology and the recipient of the 2015 Dewey Winburne Award. Karthik has previously held positions at Microsoft and Deutsche; he was invited to the White House on two occasions to present his research on the computational detection of cyberbullying and use of probabilistic graphical Bayesian models for crisis counseling. Karthik holds a Doctoral degree from Massachusetts Institute of Technology.

Open Data Science Conference