About Me

I’m a PhD candidate in the MIT Department of Brain and Cognitive Sciences, supervised by Josh Tenenbaum. My research synthesises Bayesian machine learning and cognitive science to develop learning systems which are more flexible, interpretable, and human-like. I believe this goal is best realised by models which embody explicit structured relationships between parts - compositionality, hierarchy, causality. Given this perspective, I aim to combine probabilistic programming (for its rich knowledge representation) with deep learning and evolutionary computation (for tractable search and inference).

You can contact me at lbh@mit.edu or lhewitt@protonmail.com

Machine Learning Research Projects

DreamCoder: paper (PLDI 2021)
Ellis et al.

Memoised wake-sleep: paper (UAI 2020)
Hewitt, Le, Tenenbaum

SketchAdapt: paper (ICML 2019)
Nye, Hewitt, Tenenbaum, Solar-Lezama

Inferring Structured Visual Concepts from Minimal Data: paper (CogSci 2019)
Qian, Hewitt, Tenenbaum, Levy

Variational Homoencoder: GitHubpaperoral presentation (UAI 2018)
Hewitt, Nye, Gane, Jaakkola, Tenenbaum

Bayesian Auditory Scene Analysis: websitepaperoral presentation (CogSci 2018)
Cusimano, Hewitt, Tenenbaum, McDermott



Tutorial on Bayesian Inference In Generative Models: video (CBMM, 2018)
with Maddie Cusimano

Deep Learning & Art projects: Gen Studio @ The Met (2019)Mind the Machine (2017)
with Sarah Schwettmann, Maddie Cusimano

Some old writing on AI: [thinking machines]

Predicting emojis [ :-) ]
with SwiftKey


Awards & Fellowships

2019 Walle Nauta Award for Continued Dedication to Teaching, MIT
2018 Angus MacDonald Award for Excellence in Undergraduate Teaching, MIT
2017 Angus MacDonald Award for Excellence in Undergraduate Teaching, MIT
2016 Second place finalist, Flame Challenge for science communication.
2015-2017 Henry E. Singleton (1940) Fellowship, MIT
2015 Dean’s List Commendation for Outstanding Academic Performance, UCL
2014 Vacation Bursary Award, Engineering and Physical Sciences Research Council