I want to understand how machines can learn useful representations of the world. I am also interested in discrete probability and its fundamental relation to many learning problems.

I am Josh Robinson, a PhD student at MIT CSAIL and LIDS advised by Stefanie Jegelka and Suvrit Sra. Previously I was an undergraduate in mathematics at the University of Warwick where I worked with Robert MacKay on probability theory.

- Paper on the use of shortcuts in contrastive learning accepted at NeurIPS 2021.
- reviewer award (top 10% of reviewers) at ICML 2021.
- New paper on the use of shortcuts in contrastive learning accepted at ICML 2021 workshop on self-supervised learning for reasoning and perception.
- Awarded the runner-up prize for the 2021 Two Sigma PhD fellowship.
- Paper on contrastive learning with hard negative samples accepted at ICLR 2021!
- reviewer award (top 10% of reviewers) at NeurIPS 2020.
- Paper on debiased contrastive learning accepted at NeurIPS 2020 with a spotlight!
- Paper on optimal minibatch selection at the ICML 2020 Workshop on Real World Experiment Design and Active Learning.
- Spent the Summer 2020 at Amazon Web Services working as an Applied Scientist. I worked on adding the first unsupervised functionality to AutoGluon, an open source AutoML platform.
- Paper on understanding theoretical foundations of pretraining embeddings using weak supervision accepted at ICML 2020!
- Paper on a new probabilistic model for diversity accepted at NeurIPS 2019.

Review service: NeurIPS, ICML, COLT, JMLR

Teaching: 6.867 machine learning (graduate level) – MIT fall 2021