Vikash K. Mansinghka
vkm@mit.edu
(office) MIT Room 46-5169
(lab) MIT Room 46-4066

Our minds accomplish far more than our best machine intelligence systems, automatically building flexible and remarkably useful models from our inherently ambiguous sensory experience. At the same time, our brains are built out of far slower, less reliable parts than even the earliest electronic computers. I try to identify computational principles that can help us narrow these capability, efficiency and robustness gaps, build better computational models of cognition, and engineer systems that help people infer the probable causes behind their data.

My research is based on an emerging marriage of the abstractions behind software and hardware with stochastic processes, random variables, and Bayesian inference. These randomized building blocks support the engineering of computing systems that stochastically explore alternative explanations for data, implicitly obeying the norms of Bayesian reasoning, rather than carry out deterministic, step-by-step calculations. So far, this approach has yielded new probabilistic programming technology and stochastic hardware designs, as well as practical machine learning systems that analyze high-dimensional tables and automatically produce statistically reliable predictions from raw, messy data.

This work draws heavily on and contributes to several other fields, especially computational Bayesian statistics and nonparametrics, probabilistic programming, machine learning, artificial intelligence, and computational cognitive science. It also builds on ideas from functional programming languages, computational statistical mechanics, and digital design.

I am currently a Fellow of MIT's Intelligence Initiative, where I lead the Probabilistic Computing Project (website coming soon), focused on building, analyzing and deploying a range of probabilistic computing systems. We are supported by MIT's Computer Science and Artificial Intelligence Laboratory (especially Google's "Rethinking AI" project), the Department of Brain & Cognitive Sciences, and Harvard's School of Engineering and Applied Sciences (where I am a Visiting Fellow). I am also involved in consulting and advisory relationships with industry, and co-founded a startup that was ultimately acquired by Salesforce.com in 2012. I have served on DARPA's Information Science and Technology (ISAT) advisory board, and currently serve on the Editorial Board for the Journal of Machine Learning Research. I received my PhD in 2009 from MIT, advised by Professor Joshua Tenenbaum.

Some of my publications, in reverse chronological order (last updated early 2011):

[PDF] Shafto, Kemp, Mansinghka, Tenenbaum. A probabilistic model of cross-categorization. Cognition, 2011. (Cognitive science application of cross-categorization method.)

[PDF] Freer, Mansinghka, Roy. When are probabilistic programs probably computationally tractable? NIPS 2010 Workshop on Monte Carlo Methods in Modern Applications.

[PDF] Mansinghka, Jonas, Petschulat, Cronin, Shafto, Tenenbaum. Cross-categorization: a method for discovering multiple overlapping clusterings. NIPS 2009 Workshop on Nonparametric Bayesian Statistics. (Preliminary version; journal manuscript in preparation for 2011)

[PDF] Mansinghka. Natively Probabilistic Computation. PhD Dissertation, 2009. [2009 George M. Sprowls Award for best doctoral thesis in computer science.]

[PDF] Sanborn, Mansinghka, Griffiths. A Bayesian Framework for Modeling Intuitive Dynamics. To appear in COGSCI 2009.

[PDF] Mansinghka, Roy, Jonas, Tenenbaum. Exact and Approximate Sampling by Systematic Stochastic Search. In Proceedings of the Twelfth International Conference on Artificial Intelligence and Statistics (AISTATS) 2009.

[PDF] Mansinghka, Jonas, Tenenbaum. Stochastic Digital Circuits for Probabilistic Inference. MIT CSAIL Technical Report 2008-069.

[PDF] Roy, Mansinghka, Goodman, Tenenbaum. A stochastic programming perspective on nonparametric Bayes. In the Nonparametric Bayes Workshop at ICML 2008.

[PDF] Goodman, Mansinghka, Roy, Bonawitz, Tenenbaum. Church: a language for generative models. In Proceedings of the Twenty-Fourth Conference on Uncertainty in Artificial Intelligence (UAI) 2008.

[PDF] Goodman, Mansinghka, Tenenbaum. Learning grounded causal models. In Proceedings of the Twenty-Ninth Annual Conference of the Cognitive Science Society (COGSCI) 2007. (The experiment demo is here.) [2007 Cognitive Science Society computational modeling prize for Perception and Action.]

[PDF] Frank, Goldwater, Mansinghka, Griffiths, Tenenbaum. Modeling human performance in statistical word segmentation. In Proceedings of the Twenty-Ninth Annual Meeting of the Cognitive Science Society (COGSCI) 2007.

[PDF] Mansinghka, Roy, Rifkin, Tenenbaum. AClass: An online algorithm for generative classification. In Proceedings of the 11th International Conference on Artificial Intelligence and Statistics (AISTATS) 2007.
[PDF] Roy, Kemp, Mansinghka, Tenenbaum. Learning annotated hierarchies from relational data. In Neural Information Processing Systems (NIPS) 19. December, 2006.
[PDF] Mansinghka, Kemp, Tenenbaum, Griffiths. Structured priors for structure learning. In Proceedings of the Twenty-Second Conference on Uncertainty in Artificial Intelligence (UAI) 2006.
[PDF] Shafto, Kemp, Mansinghka, Gordon, Tenenbaum. Learning cross-cutting systems of categories. In Proceedings of the Twenty-Eighth Annual Conference of the Cognitive Science Society (COGSCI) 2006.
[PDF] Goodman, Bonawitz, Baker, Mansinghka, Gopnik, Wellman, Schulz, Tenenbaum. Intuitive theories of mind: a rational approach to false belief. In Proceedings of the Twenty-Eighth Annual Conference of the Cognitive Science Society (COGSCI) 2006.