I am a member of technical staff at OpenAI. I did my PhD at UC Berkeley EECS as a member of Berkeley AI Research, where I was fortunate to be advised by Alexei A. Efros and funded by the PD Soros Fellowship. My main interests are in scalable objectives and architectures for self-supervised and unsupervised learning.

Previously, I've spent time as an intern at DeepMind in London and a student researcher at Google Brain.
Before grad school, I was a Research Engineer at Facebook AI Research in New York and studied Computer Science at Princeton (B.S. 2015).


Teaching

CS 189/289: Introduction to Machine Learning (Spr 2022, Grad TA)
CS 280: Graduate Computer Vision (Spr 2019, Lead TA)


Selected Work (scholar page)


Scalable Adaptive Computation for Iterative Generation.
ArXiv, In Submission.
A Jabri, D Fleet, T Chen.

Universal neural architecture that can adaptively allocate capacity and computation for iterative generation of high-dimensional data. State-of-the-art image and video generation with pure attention-based architecture.
[ paper ] [ project page (soon>) ] [ code (soon) ]
Space-Time Correspondence as a Contrastive Random Walk.
NeurIPS 2020, Oral Presentation.
A Jabri, A Owens, A Efros.

Dense representation learning from unlabeled video, by learning to walk on a space-time graph.
[ paper ] [ project page ] [ code ]
Unsupervised Curricula for Visual Meta-Reinforcement Learning.
NeurIPS 2019, Spotlight Presentation.
A Jabri, K Hsu, B Eysenbach, A Gupta, S Levine, C Finn.

Unsupervised discovery and meta-learning of visuomotor skills, by deep clustering your own trajectories.
[ paper ][ project page ]
Towards Practical Multi-Object Manipulation using Relational Reinforcement Learning.
ICRA 2020.
R Li, A Jabri, T Darrell, P Agrawal.

Training a graph neural net policy with a simple curriculum leads to task decomposition that generalizes to new configurations.
[ paper ][ project page ][ code ]
Learning Correspondence from the Cycle-Consistency of Time.
CVPR 2019, Oral Presentation.
X Wang*, A Jabri*, A Efros.

Learn a generic representation for visual correspondence from unlabeled video, using cycle consistency in time.
[ paper ][ project page ][ code ]
Universal Planning Networks.
ICML 2018.
A Srinivas, A Jabri, P Abbeel, S Levine, C Finn.

Learn a visual representation that captures task semantics by differentiating through model-based planning.
[ paper ] [ project page ] [ code ]
CommAI: Evaluating the first steps towards a useful general AI.
ICLR 2017 Workshop

M Baroni, A Joulin, A Jabri, G Kruszewski, A Lazaridou, K Simonic, T Mikolov.
A short paper on the nature of tasks we are studying in the CommAI project.
Learning Visual N-Grams from Web Data.
ICCV 2017.
A Li, A Jabri, A Joulin, L van der Maaten.

A smoothed n-gram loss for learning visual representations from compositional phrases, at scale.
[ paper ]
Revisiting Visual Question Answering Baselines.
ECCV 2016.
A Jabri, A Joulin, L van der Maaten.

SOTA VQA models may not be learning what we think they are... #datasetbias
[ paper ]
Learning Visual Features from Large Weakly Supervised Data.
ECCV 2016.
A Joulin, L van der Maaten, A Jabri, N Vasilache.

Learn strong visual features from tons of hashtag data, with interesting byproducts like translation by visual grounding.
[ paper ]

ajabri at gmail