Arkabandhu Chowdhury

arjia.001 AT gmail.com

I am a Research Scientist at Meta's Super Intelligence Lab, where I work on large-scale multimodal understanding. I built 0-to-1 video pretraining infrastructure for Meta's frontier MSL models, including dataloaders for diverse video datasets and video evaluation infrastructure, powering Muse Spark. Previously, I led video pretraining for Llama 4 and contributed to image post-training for Llama 3.

Prior to that, I worked on vision understanding for Instagram and Facebook recommendation systems at Meta, and on natural language understanding for Alexa AI at Amazon. I earned my M.S. and Ph.D. in Computer Science from Rice University, advised by Prof. Chris Jermaine, and my B.E. in Electronics and Telecommunication Engineering from Jadavpur University. I have also interned or worked at IBM Research, Anadarko Petroleum, and Hewlett-Packard.

Publications [Full List]

* indicates equal contribution

Llama 3

The Llama 3 Herd of Models

Meta Llama team

CALVIN

CALVIN: Improved Contextual Video Captioning via Instruction Tuning

G. Somepalli, A. Chowdhury, R. Basri, J. Geiping, T. Goldstein, D. Jacobs

NeurIPS 2024 | Paper

Hiera

Hiera: A Hierarchical Vision Transformer without the Bells-and-Whistles

C. Ryali et al. (multiple authors from Meta AI/FAIR)

Meta-Meta Classification

Meta-Meta-Classification for One-Shot Learning

A. Chowdhury, D. Chaudhari, S. Chaudhuri, C. Jermaine

Few-shot Learning

Few-shot Image Classification: Just Use a Library of Pre-trained Feature Extractors and a Simple Classifier

A. Chowdhury, M. Jiang, S. Chaudhuri, C. Jermaine

Multimodal Dialog

Multimodal dialog for browsing large visual catalogs using exploration-exploitation paradigm in a joint embedding space

A. Chowdhury*, I. Bhattacharyya*, V. Raykar

MCMC

Parallel and Distributed MCMC via Shepherding Distributions

A. Chowdhury, C. Jermaine

Academic Service

Conference & Journal Reviewer

ICML, NeurIPS, CVPR, ICCV, ECCV, WACV, CIMR, ACMMM, AMLC, ML4H, JMLC, etc.