Bram Wallace
Research Scientist at Salesforce (Generative AI)
Email: bw462@cornell.edu
Hi! I’m Bram Wallace, a research scientist in Salesforce’s Generative AI team.
My research focuses on image generation and editing. My favorite approach to research is to uncover hidden capabilities and properties of pretrained neural networks and algorithmically exploit these insights, or leverage them to train even better neural networks. An example of my recent work is EDICT (Exact Diffusion Inversion via Coupled Transformations) where we design an algorithm that modifies the generative process of diffusion models to be exactly invertible. We leverage this work for detail-preserving image-editing and will be releasing follow-ups soon. See my Publications tab or Google Scholar for past work.
While at Salesforce, I’ve also worked on language modeling of amino acid sequences for novel protein generation and written an internal library for large-scale training of vision-language models on TPUs using Pytorch/XLA.
I completed my PhD at Cornell advised by Dr. Bharath Hariharan performing research at the intersection of self-supervised and transfer learning. My undergraduate studies were at the College of San Mateo and University of California: Santa Barbara. During the two-year gap between UCSB and Cornell, I performed research with Dr. Paul Atzberger (see my Publications section) and did a year-long internship at the biotechnology startup Vium. In addition to my research at Cornell during my PhD, I’ve also worked in the Financial Engineering team at Bloomberg LP and interned at Salesforce Research.
In my free time I enjoy trail running, hiking, backpacking, river tubing, lifting, board/card games, reading, and disc golf.