Soumith Chintala

Open-source work

Soumith Chintala co-founded PyTorch and earlier maintained Torch-7. Before 2012, Chintala helped maintain EBLearn, a C++ deep learning framework, with Pierre Sermanet and Yann LeCun, and later created convnet-benchmarks, a deep learning benchmarking suite that hardware makers used from 2015 to 2017 to tune their systems.¹

Chintala is the last-listed author of the 2019 paper "PyTorch: An Imperative Style, High-Performance Deep Learning Library." PyTorch's install page offers builds for NVIDIA CUDA, AMD ROCm and CPU-only machines, and ExecuTorch extends the project to running models on phones and other devices.⁴²³

Research

Chintala co-authored three papers on generative adversarial networks: LAPGAN (2015), which generates images with a Laplacian pyramid of adversarial networks, DCGAN (2015), on unsupervised representation learning with deep convolutional GANs, and Wasserstein GAN (2017). Other research has covered object and human detection, generative modeling of images and video, AI for video games and machine learning systems.¹⁵⁶⁷

At NYU, Chintala works on home robotics with Lerrel Pinto, often using a Hello Robot Stretch. Their papers include "On Bringing Robots Home" (2023) and "Robot Utility Models" (2024), which studies general policies that can be deployed zero-shot in new environments.¹⁸⁹

Career

Chintala's homepage lists 11 years at Meta and current work at Thinking Machines and NYU on AI infrastructure, AI research and robotics.¹

Work and contributions

• PyTorch

  Co-founded PyTorch. Its local install page offers builds for NVIDIA CUDA, AMD ROCm, and CPU-only machines.¹²

• ExecuTorch

  ExecuTorch, which its README describes as PyTorch's open-source stack for running AI locally on phones and other devices, is part of the PyTorch project Chintala co-founded.¹³

• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks (DCGAN) Published research

  Co-author of the 2015 DCGAN paper, one of three generative adversarial network papers, with LAPGAN and Wasserstein GAN, that Chintala's homepage lists among Chintala's research.⁶¹

• PyTorch: An Imperative Style, High-Performance Deep Learning Library Published research

  Last-listed author of the 2019 paper describing PyTorch's design and implementation.⁴

Timeline

1. Jun 18, 2015Co-authored LAPGAN, deep generative image models using a Laplacian pyramid of adversarial networks.⁵
2. Nov 19, 2015Co-authored the DCGAN paper on deep convolutional generative adversarial networks.⁶
3. Jan 26, 2017Co-authored the Wasserstein GAN paper.⁷
4. Dec 3, 2019Co-authored the paper describing PyTorch as an imperative-style, high-performance deep learning library.⁴
5. Nov 27, 2023Co-authored "On Bringing Robots Home" with NYU collaborators.⁸
6. Sep 9, 2024Co-authored "Robot Utility Models" on zero-shot deployment of robot policies in new environments.⁹

Other documented roles

Roles the sources document that are not confirmed as current.

• 11 years at Meta, Meta¹

Where they work

• Thinking Machines Lab 13 people in the wiki
• NYU 12 people in the wiki

Related people

• Nikita ShulgaAlso at Thinking Machines Lab
• Sam GrossAlso at Thinking Machines Lab
• Horace HeAlso at Thinking Machines Lab
• Liliang RenAlso at Thinking Machines Lab
• Woosuk KwonAlso at Thinking Machines Lab
• Nikhil ThoratML frameworks and libraries
• Abdul Fatir AnsariML frameworks and libraries
• Sayak PaulML frameworks and libraries
• Lysandre DebutML frameworks and libraries
• Johannes GäßlerML frameworks and libraries