JP Chen
Generative AI Researcher
me [@] jonathanpchen.com
I am the cofounder of a stealth startup in the AI x Gaming space, working on scaling the performance and reasoning capabilities of local, offline language models. Previously I was at Meta, where
I worked on generative models on the computer vision team at Instagram.
My research interests are broadly at the intersection of interpretability and logical reasoning of large language models on a theoretical, foundational level.
In a previous life, I developed probabilistic programming languages and was the founding coauthor of both Bean Machine
and Pyro. While at Uber AI Labs, I worked on research in
approximate inference, Bayesian nonparametrics, and sensor fusion for self driving cars.
I received my undergraduate degree in computer science from the University of Pennsylvania.
I am a Varsity and JV volleyball coach at Summit Shasta High School in my spare time. I occasionally write about things I find interesting.
Papers
Preprints
Evaluating the generative models produced by language models
Jonathan P. Chen
Preprint.
Automatic Model Construction for Daily Time Series Forecasting
Zhen Cao*, Jonathan P. Chen*, Neeraj Pradhan.
Preprint.
*Equal Contribution.
Inverse Graphics for Transfer Learning of Street Signs
Jonathan P. Chen, Fritz Obermeyer, Paul Szerlip.
Preprint.
TreeCat: a Bayesian Latent Tree Model of Sparse Heterogeneous Tabular Data
Fritz Obermeyer, Jonathan P. Chen, Martin Jankowiak.
Preprint.
Publications
Joint Mapping and Calibration via Differentiable Sensor Fusion
Jonathan P. Chen*, Fritz Obermeyer*, Vlad Lyapunov, Lionel Gueguen, Noah Goodman.
NeurIPS ML4AD Workshop.
*Equal Contribution.
Functional Tensors for Probabilistic Programming
Fritz Obermeyer, Eli Bingham, Martin Jankowiak, Du Phan, Jonathan P. Chen.
NeurIPS Program Transformations Workshop.
Generating experimentally unrelated target molecule-binding highly functionalized nucleic-acid polymers using machine learning
Jonathan C. Chen, Jonathan P. Chen, Michael Wornow, Minwoo Bae, Wei-Hsi Yeh, Adrian Berliner, David R. Liu.
Nature Communications.
Improving Automated Variational Inference with Normalizing Flows
Stefan Webb, Jonathan P. Chen, Martin Jankowiak, Noah D. Goodman.
ICML AutoML Workshop, 2019.
Pyro: Deep Universal Probabilistic Programming
Eli Bingham, Jonathan P. Chen, Martin Jankowiak, Neeraj Pradhan,
Theofanis Karaletsos, Rohit Singh, Paul Szerlip, Paul Horsfall, Noah D. Goodman.
Journal of Machine Learning Research, 2018.
Transpiling Stan models to Pyro
Jonathan P. Chen, Rohit Singh, Eli Bingham, Noah Goodman.
The International Conference on Probabilistic Programming (PROBPROG), 2018.
Open Source Software
Pyro
Pyro is a deep, universal probabilistic programming language written in Python on top of PyTorch, specializing in Variational Inference.
For more information, check out the release blog.
Bean Machine
Bean Machine is a flexible, declarative probabilistic programming language written on PyTorch and a custom C++ backend, specializing in MCMC Inference.
For more information, check out the release blog.
Time Series Autoforecaster
Automatic Gaussian Process kernel discovery for time series data.
Pyro-Stan Compiler
C++ Compiler that compiles Stan models into Pyro models. Validated on a corpus of 130+ models.
Torch JS
Implementation of adnn, a Javascript tensor and reverse-mode automatic differentiation library in Torch.
Patents
System and Method for Object Location Detection from Imagery
Fritz Obermeyer, Jonathan P. Chen, Vladimir Lyapunov,
Lionel Gueguen, Noah Goodman, Ben Kadlec, Douglas Bemis.
US Patent, 16/536,869. 2019.