We aim to improve our ability to engineer artificial intelligence, reverse-engineer natural intelligence, and deploy applications that increase our collective intelligence and well-being.
Our work integrates probabilistic inference, generative models, and Monte Carlo methods into the building blocks of software, hardware, and other computational systems. For example, we have developed high-level probabilistic programming languages, automated Bayesian data modeling systems, Bayesian inverse graphics approaches to 3D computer vision, and near-optimal algorithms, circuits, and hardware architectures for Monte Carlo. We test our results by collaborating with domain experts on practical applications.
In addition to core academic research and teaching, we mentor engineers and entrepreneurs, develop open source software, and lead hands-on AI workshops for industry. These activities have led to: VC-backed startups acquired by Salesforce (2012) and Tableau (2018), the founding in 2020 of Common Sense Machines, and a new Intel Center for Probabilistic Computing. We also carry out joint development and field testing with partners from both the public and private sectors.
Contact us to get involved in testing or contributing
Latest News
-
May 2023: Congratulations to Dr. Feras Saad, for winning the George M. Sprowls PhD Thesis Award in Artificial Intelligence and Decision-Making for his thesis “Scalable Structure Learning, Inference, and Analysis with Probabilistic Programs”.
-
January 2023: Our research paper SMCP3: SMC with Probabilistic Program Proposals has been accepted for publication at AISTATS 2023.
-
January 2023: Our research paper ADEV: Sound Automatic Differentiation of Expected Values of Probabilistic Programs has been presented at POPL 2023 and covered by MIT News.
-
November 2022: Our research paper Visual object detection biases escape trajectories following acoustic startle in larval zebrafish has been accepted for publication at current biology and covered by MIT News.
-
April 2022: Our research paper Estimators of Entropy and Information via Inference in Probabilistic Models has been presented at AISTATS 2022 and covered by MIT News.
-
October 2021: Our research on 3DP3: 3D Scene Perception via Probabilistic Programming has been accepted for publication at NeurIPS 2021 and covered by MIT News.
-
August 2021: Our research on probabilistic programming with fast exact symbolic inference was covered by MIT News
-
May 2021: Our research on Bayesian Data Cleaning at Scale with Domain-Specific Probabilistic Programming was covered by MIT News
-
May 2021: We are happy to share that our paper Hierarchical Infinite Relational Model has been accepted to for publication and oral (long-form) presentation at UAI 2021.
-
March 2021: We are pleased to announce that our paper SPPL: Probabilistic Programming with Fast Exact Symbolic Inference has been accepted to PLDI 2021.
-
February 2021: We are happy to announce that our paper PClean: Bayesian Data Cleaning at Scale with Domain-Specific Probabilistic Programming has been accepted for publication and oral presentation at AISTATS 2021.
-
We are pleased to share that Vikash Mansinghka is an invited keynote speaker at LAFI 2021.
-
January 2021: We are happy to announce that our paper Transforming Worlds: Automated Involutive MCMC for Open-Universe Probabilistic Models has been accepted at the 3rd Symposium on Advances in Approximate Bayesian Inference (AABI).
-
December 2020: Our research on Online Bayesian goal inference for boundedly-rational planning agents was covered by MIT News
-
December 2020: We are pleased to share that we had two papers accepted at LAFI 2021:
-
On the Automatic Derivation of Importance Samplers from Pairs of Probabilistic Programs. Alexander K. Lew, Ben Sherman, Marco Cusumano-Towner, Michael Carbin, Vikash Mansinghka
-
Genify.jl: Transforming Julia into Gen to enable programmable inference. Tan Zhi-Xuan, McCoy R. Becker, Vikash Mansinghka
-
-
September 2020: We are happy to announce that our paper Online Bayesian goal inference for boundedly-rational planning agents has been accepted for publication at NeurIPS 2020.
-
June 2020: We are pleased to share that our paper Causal inference using Gaussian processes with structured latent confounders has been accepted for publication at ICML 2020.
-
May 2020: Our research on The Fast Loaded Dice Roller was covered by MIT News, Intel, and Quanta.
-
January 2020: We are happy to announce that our paper The Fast Loaded Dice Roller: A Near-Optimal Exact Sampler for Discrete Probability Distributions has been accepted for publication at AISTATS 2020.
-
November 2019: We are pleased to announce that two papers have been accepted to POPL 2020:
-
A Type System and Semantics for Sound Programmable Inference in Probabilistic Languages. Alexander K. Lew, Marco F. Cusumano Towner, Benjamin Sherman, Michael Carbin, Vikash K. Mansinghka.
-
Optimal Approximate Sampling from Discrete Probability Distributions. Feras A. Saad, Cameron E. Freer, Martin C. Rinard, Vikash K. Mansinghka.
-
-
June 2019: Our research on Gen was covered on MIT News and further covered by VentureBeat, ZDNet, and featured on Hacker News.
-
April 2019: We are happy to announce that our paper Gen: A General-Purpose Probabilistic Programming System with Programmable Inference was accepted to PLDI 2019.
-
January 2019: Our research on Bayesian Synthesis of Probabilistic Programs for Automatic Data Modeling, presented at POPL 2019, was featured on MIT News and ZDNet.
-
January 2019: We are happy to announce that our paper A Family of Exact, Distribution-Free Goodness-of-Fit Tests for High-Dimensional Discrete Distributions was accepted to AISTATS 2019.
-
November 2018: We are pleased to announce that our paper Bayesian Synthesis of Probabilistic Programs for Automatic Data Modeling was accepted to POPL 2019. See here for a video of the talk presented at the conference.
-
October 2018: We hosted the Inaugural International Conference on Probabilistic Programming (PROBPROG 2018) organized by Vikash Mansinghka, Jan-Willem van de Meent, and Avi Pfeffer. Lab members Marco Cusumano-Towner, Feras Saad, and Ulrich Schaechtle presented research talks and Vikash Mansinghka hosted a tutorial session. Videos of talks are available here.
-
August 2018: We gave an invited talk at the Joint Statistical Meetings (JSM) 2018 on Probabilistic Programming with Non-Parametric Bayesian Model Discovery in BayesDB.
-
June 2018: Tableau acquired Empirical, the second startup based on components of the MIT Probabilistic Computing Stack, into their ‘automated intelligence’ products group, to provide automated data insights.
-
June 2018: The Defense Advanced Research Projects Agency (DARPA) selected Ulrich Schaechtle, post-doctoral researcher, as a DARPA Riser for 2018. Dr. Schaechtle will attend and present on machine-assisted causal modeling via probabilistic programming at DARPA’s 60th Anniversary Symposium (D60) in September at the Gaylord National Convention Center in National Harbor, Maryland.
-
May 2018: We gave an invited talk at the AI for Good Global Summit held by the UN International Telecommunications Union
-
April 2018: We are presenting lab member Dr. Ulrich Schechtle’s work on AI-assisted data search for collaborative science, using BayesDB, at the upcoming DARPA SD2 Working Meeting in San Jose, CA. This ongoing project is building on our pre-print Probabilistic search for structured data via probabilistic programming and nonparametric Bayes.
-
April 2018: Our workshop paper titled Gen: a probabilistic programming platform with fast custom inference via code generation was accepted to MAPL 2018 (Machine Learning and Programming Langauges Workshop), affiliated with PLDI 2018. Contact PhD student Marco-Cusumano-Towner for more information on Gen.
-
April 2018: We gave an invited keynote on AI-assisted data science with BayesDB at Deloitte’s annual Data Science Summit in Dallas, Texas, to 200+ data scientists and business leaders from across the organization.
-
March 2018: We gave an invited keynote on AI-assisted data science at the Alan Turing Institute’s Workshop on Artificial Intelligence for Data Analytics. A key theme we discussed was how probabilistic programming can help the open science and data science communities develop automated tools for searching data, screening data for quality errors, and helping to design more efficient experiments.
-
March 2018: We had two papers accepted to PLDI 2018: Probabilistic Programming with Programmable Inference and Incremental Inference for Probabilistic Programming.
-
March 2018: We gave an invited presentation on BayesDB for AI-assisted data search at the Investigative Reporters and Editors’ 2018 NICAR conference on computer assisted reporting. We are excited to be collaborating with leading newsrooms on data search and data screening; for more information, please contact lab member Sara Rendtorff-Smith.
-
January 2018: Our paper A Bayesian Nonparametric Method for Clustering, Imputation, and Forecasting in Multivariate Time Series was accepted to AISTATS 2018.
-
January 2018: We presented Using probabilistic programs as proposals at the Probabilistic Programming Semantics (PPS) Workshop 2018, co-located with POPL.