Prateek Verma

Thank you for visiting my webpage. I am currently a researcher at Stanford University. I am interested in the following questions.

  • How to advance creativity of computers w.r.t audio, text, music, video generation.

  • How do we understand and advance our perception of the world via modern advancements? (text, videos, audio, music)

  • Doing classic signal processing research with all of the fancy tools/toys available to us these days.

  • Advance Large Language Models Across Modalities.

    • How to make pretraining faster upton 50% by using inductive biases. We have a paper coming up soon.

    • How to incorporate really really long context >1M tokens

    • How to utilize internal circuitary of these models

My goal recently is also to make computers hear with the best possible ears surpassing what evolution gave us. We have an ICASSP submission that allowed us to switch learned machine ears every 25ms depending upon the contents of input waveform.

Some of my research on LLMs and Transformers featured on Stanford's YouTube channel, with over 5.1k views (https://www.youtube.com/watch?v=wvE2n8u3drA). In addition, I have been fortunate that Stanford Human-Centered Insitute for Artificial Intelligence has supported my work through generous Google cloud computing grants.

Bio: I have a strong interdisciplinary background and approach my research like a bird hopping in different fields like Natural Language Processing, Computer Vision, Signal Processing, Machine learning, and Robotics. I did my AI Residency at Google X, working on an undisclosed project(s). Before Stanford, I was at the Indian Institute of Technology (IIT) Bombay, the top-most school in India, graduating with an M. Tech in Electrical Engineering with a specialization in Signal Processing. I like running and biking. I actively played field hockey, cricket, table-tennis, during IIT Bombay days. I listen to all kinds of music and am awe in the genius of Rohail Hyatt and Trent Reznor-Atticus Ross weaving such magical compositions.

While I am not on academic twitter/linkedin, I'm happy to connect via email! Drop me a line at “id@stanford.edu” where id is prateekv