Sarthak Kamat

sartk at berkeley dot edu

I'm a research resident at Berkeley AI Research advised by Trevor Darrell and Ilija Radosavovic. I'm working on learning motor control for humanoid robots with transformers.

I studied Computer Science at UC Berkeley from 2020 - 2023, where I graduated with highest distinction.

Research

• Humanoid Locomotion as Next Token Prediction
  Ilija Radosavovic, Bike Zhang, Baifeng Shi, Jathushan Rajasegaran,
  Sarthak Kamat, Trevor Darrell, Koushil Sreenath, Jitendra Malik
  arXiv website

• Revisiting Generalizability in Deepfake Detection
  Sarthak Kamat, Shruti Agarwal, Trevor Darrell, Anna Rohrbach
  ICCV 2023 oral

• Surface Spherical Encoding and Contrastive Learning for Virtual Bone Shape Aging
  Francesco Caliva*, Sarthak Kamat*, Alejandro Morales Martinez, Sharmila Majumdar, Valentina Pedoia
  MIDL 2021 special issue

Projects

• In-Hand Cube Reorientation
  I trained an RL policy for in-hand cube reorientation with the Allegro hand, in the IsaacGym simulator. The model gets RGB pixels in, and outputs motor actions.

• Raytracing Schwarzschild Black Holes
  Aseem Doriwala, Sarthak Kamat, Vincent Lim, Jeremy Ferguson
  CS 184: Computer Graphics, Project Showcase Winner, 2023
  website showcase

• NP-Hard Combinatorial Optimization
  Araav Patel, Sarthak Kamat, Vincent Lim
  4th place out of 180, CS 170 Fall 2021
  leaderboard writeup

Other Experience

• Hudson River Trading
  Algorithm Developer Intern (summer 2023)
  I trained low-latency models for automated trading, and wrote algorithms for arbitrage detection in C++.

• Berkeley Vision Lab
  Deepfake Detection Researcher (summer 2022 - spring 2023)
  I worked on building generalizable deepfake detection algorithms with Anna Rohrbach and Shruti Agarwal.

• UCSF Center for Intelligent Imaging
  Computer Vision Researcher (summer 2021)
  I trained a 3D generative model for virtual bone shape aging to model the onset of osteoarthritis. I was mentored by Francesco Caliva and Valentina Pedoia.

Teaching

• Course Assistant and Reader for EECS 127 (Optimization Models in Engineering), Spring 2022.
• Academic Intern for CS 61A in Fall 2022, the intro CS class at UC Berkeley.
• I taught machine learning to middle schoolers at public libraries in San Jose with Sahil Jain.

Misc

• Africa by Toto on the electric guitar.