I try to keep this page updated with teasers of what I’m currently working on. Please reach out to me if you’re interested in specifics or possible collaborations.
Self-Supervised Ensembling: My work at Salesforce Research focused on the problem of self-supervised ensembling of pretrained models in the transfer setting. This work will be submitted for review in October.
Satellite Imagery Time Series: Our group is working along with crop scientists at Cornell on the problem of crop species classification in satellite imagery. Our focus is on the problem of generalizing across regions. As part of this research, we’ve developed new techniques for large-scale timeseries deep metric learning. This work will be submitted for review in October.
Expert Model Selection: Task2Vec is one of my favorite papers to come out in recent years. We developed self-supervised versions of this method that don’t require target labels to select the proper expert, as well as versions of the algorithm which employ Neural Tangent Kernel theory to alleviate the need for a well-trained probe network. See this work in CVPR 2021!