Sanjukta Bhattacharya; Values Blog

Now I'm based at the Universtity of Bristol as an AI phd student. Prior to this, I was based in Los alamos, New mexico,as an Independent researcher at LANL.

I've been training models to learn gene expression data and how it can be used to predict cell fate.

Discrete diffusion models for gene expression data we wanted to learn to generate gene expression data in the raw-count space or the 'pixel space'. Existing methods don't work. We tried out a fundamentally different arch in this domain.

I find the idea of using gene-expression vocab based FMs/"language" models to predict counterfactual biological trajectories very nuanced!

Before this, I was working on reconstructing AI architectures at information-theoretic bounds for sample optimality. This work is motivated by the limited learning efficiency of the current LLM stack.

At the moment, I'm trying out these ideas on Modern Hopfield Networks.

I completed a Master's by Research at the University of Edinburgh, under the supervision of Prof. Antonio Vergari. My thesis was based on compilation and approximation of probabilistic circuits from Markov random fields.

Email  /  Twitter  /  Github  /  Linkedin