Ria Vinod

About me

I'm currently a 3rd Year PhD student at Brown University, advised by Lorin Crawford. I work on biologically and physically inspired generative models, with a focus on learning meaningful, data-efficient representations for protein-design problems.

Before that, I studied Electrical Engineering and Computer Science at UC Berkeley and graduated in 2021.

I've spent summers at IBM Research, Salesforce and Optum, working on different research and data engineering problems. Most recently, I was a Machine Learning intern at Prescient Design, where I developed principled score matching methods for biological manifolds, applied to sequence design.

I'm excited about the future of Machine Learning for scientific domains,and its applications in climate, medicine and synthetic life. My reading and work leans into the mathematical and algorithmic principles that we can use to explain structure and mechanisms in science. As I learn more about math and proteins, I'm trying to make notes and collect useful links. You can find notes here and interesting reads here. Some of the workflows that I've created as I've needed them can be found on my github.

I also spend time co-organizing a Machine Learning x Proteins seminar series and co-run after-school computational science workshops at Providence public schools with some of my wonderful peers.

As a new graduate student, I thought these resources were super useful (some Boston-area specific ones too!).

Research Projects

• Logit Subspace Diffusion for Protein Sequence Design.
  With Nathan Frey, Andrew Watkins,and Jae Hyeyeon Lee.
  Developed a score-based stochastic differential equation (SDE) framework that diffuses over zero-identity component simplexes of protein sequences for antibody design.
  
  
• Diffusion Models for Sequence-Structure Co-Design. arxiv
  With Kevin Yang and Lorin Crawford.
  Investigating sequence-structure distribution correlation through diffusion-based generative modelling for various downstream protein prediction tasks.
  Presented at LMRL @ NeurIPS 2022. Spotlight.
  
  
• Reprogramming Pretrained Language Models for Protein Sequence Representation Learning. arxiv
  With Pin-Yu Chen and Payel Das.
  Introduces a representation learning framework in which large language models are reprogrammed for alternate tasks that can perform well in low-training data settings.
  In review at Nature Biomedical Engineering.
  
  
• Representation Learning for Molecular Property Prediction. arxiv
  With Pin-Yu Chen and Payel Das.
  Introduces dictionary learning to utilize learned representations from pretrained deep models for functional and structural molecular property prediction.
  Presented at WiML, LMRL @ NeurIPS 2020.