Anshuman Sinha

Hello! My name is Anshuman. I am currently a Generative AI Engineer at Apple, where I work on Vision-Language Action Models, Imitation Learning, and Reinforcement Learning.

My interests lie at the intersection of Generative AI and Science, with a focus on developing intelligent systems that can learn from and reason about complex multimodal environments. I’m particularly excited about building Bayesian learning architectures with physics-based priors for applications in natural sciences.

Previously, I worked with the Fung Group at Georgia Tech under Prof. Victor Fung, where I explored Deep Generative Models for Scientific Discovery and contributed to research on Multimodal Language Models.

I hold a Master’s degree in CSE from Georgia Tech, and a B.Tech in Materials Science and Computation from IIT Kanpur, where I was awarded the Proficiency Gold Medal.

My broader research interests include LLMs, Latent Variable Models, and their applications in Natural Sciences and for Social Good. Feel free to reach out if you’d like to connect or collaborate. On a side note, if you're interested in initiatives aimed at educating underprivileged children, I’d love to hear from you.

Recent Research

Generative Diffusion model for structure representation

Anshuman Sinha, Shuyi Jia, Victor Fung,

Under open-review

ICLR 2024 (prospective)

Soon   |   Paper   |   Project page   |   Code   |  

Multi-modal LLMs: Instilling time through post-pretraining

Anshuman Sinha, Camille Migozzi, Aubin Rey ,

Under open-review

ICML 2024 (prospective)

Soon   |   Paper   |   Project page   |   Code   |  

Generative AI: Calorimer design SSL

LLNL and LBNL authouths/u>,

In draft

Internship projects

ArXiv   |   Paper   |   Project page   |   Code   |  

SSL: Graph SSL

Anshuman Sinha, Bhavay Aggrawal

MLC 2023

Course project on Self-supervised graph learning

Submission   |   Code   |   PPT  

Vision generative model: Images DDPM

Anshuman Sinha (Reproducibility)

from lucidrain

Lucid rain's implementation for my research

Code   |   PPT (⌛)