Recent Highlights

• [Sep, 2024] Our paper Superposed Decoding: Multiple Generations from a Single Autoregressive Inference Pass has been accepted at NeurIPS 2024!
• [Sep, 2023] Our paper Are “Hierarchical” Visual Representations Hierarchical? has been accepted to the Workshop on Symmetry and Geometry in Neural Representations at NeurIPS 2023.

Selected Publications

• 

  Superposed Decoding: Multiple Generations from a Single Autoregressive Inference Pass

  Ethan Shen, Alan Fan, Sarah M Pratt, Jae Sung Park, Matthew Wallingford, Sham M. Kakade, Ari Holtzman, Ranjay Krishna, Ali Farhadi, Aditya Kusupati

  NeurIPS 2024

  [PDF] [Code]

  

  Are "Hierarchical" Visual Representations Hierarchical?

  Ethan Shen, Ali Farhadi, Aditya Kusupati

  Symmetry and Geometry in Neural Representations, NeurIPS 2023

  [PDF] [Code]

Projects

Video-Language Models for Out-of-the-Box Reward

I experimented with Video-Language Models as drop-in replacements for reward functions during training on environments such as MountainCar and CartPole.

Code

Generative Visual Question Answering

I worked with two classmates to create an experimental Visual Question Answering dataset with future shifted images generated using Stable Diffusion 2.0.

PDF

Mutorials

I helped found and develop Mutorials, an ongoing online science trainer to make SAT and AP test preparation fun and accessible.

Code Site

History

• Gregory of Tours and Einhard: Clovis and Charlemagne

  I compare the credibility and accuracy of Einhard and Gregory of Tours as sources for Clovis and Charlemagne. [PDF]

  The Detroit Riots of 1967

  I contend that the Detroit Riots of 1967 were primarily caused by the decentralization of Detroit's automobile industry in the decade prior. [PDF]