Postdoc Position in Expanding Machine Learning Beyond Social Prediction to Explanation, Intervention

August 2nd, 2018 in

Title: Postdoctoral Position in Expanding Machine Learning Beyond Social Prediction to Explanation & Intervention @ the Knowledge Lab, UChicago

The Knowledge Lab at the University of Chicago seeks to hire outstanding candidates for a postdoctoral research project with support from DARPA to extend the limits of machine learning from the prediction of social systems to explaining those systems and intervening in them. This is in association with the “Ground Truth” program at DARPA. Other teams will generate reasonable agent-based models of diverse social systems, and our task is to build automated, analytical techniques that induce the "ground truth" or structure of the model and program used to generate them. We will also predict future instances of these social systems, and propose desirable and pragmatic interventions in them. Our team, the "Social MIND (Machine Inference for Novel Discovery)”, is exploring approaches that use large-scale Bayesian inference, probabilistic programming, deep learning neural networks, and approaches that link them together. We are recruiting for 1-2 postdoc positions at the intersection of data science, machine learning, automated scientific discovery, and social science.  

Postdoctoral candidates will design and conduct independent research, in close collaboration with UChicago professor, Santa Fe Institute external faculty member, and Director of Knowledge Lab James Evans, along with Joshua Tenenbaum, computational cognitive scientist from MIT, and Michael Franklin, a computer scientist and leader in systems design at the University of Chicago. Candidates must hold a PhD in Computer Science, or have substantial computational and data science background and a Ph.D. in Statistics, Applied Math, Physics, Sociology or another Social or Behavioral Science, or a related field. Candidates should have a strong publishing record. Experience in a social science discipline a strong plus. Comfort working collaboratively with a team is essential.

Specifically, successful candidates will be responsible for generating and automatically decoding agent-models, and applying these techniques to real social systems. Experience with some of the following will be helpful: deep neural networks, Bayesian inference, probabilistic programming, machine learning and machine understanding. Candidates will be involved in both innovating new methods for specific inference tasks, and assembling approaches into automated data analytic pipelines. The broader project will also involve crowdsourcing alternative approaches, so experience with crowdsourcing and intelligent model combination also a plus. Because we will be requesting social data from the agent-modeling teams, understanding social science data gathering methods and familiarity with game theoretic and agent-based models will be very helpful.  

To apply, please send CV and names for letters from at least two references to Candice Lewis,