Development of an Open-Source EEG Foundation Model using Diffusion

[alia-abbas]

My name is Ayan Ansari, I’m a senior at Greenville Tech College. I study applied science and biology and am interested in pursing medical school soon.

I’ve previously worked on machine learning surrounding computational structural biology under Dr. Guillermo Goldstein at Georgia Tech.

I was initially interested in clinical investigation after joining the Yale Center for Clinical Investigation in 2021 as a research intern. During the summer, I coordinated a team to promote pediatric research at Yale New Haven Health System, working on exposure to MyChart and software changes.

Recently, I just wrapped a position up at Dr. Ethan Kung’s lab at Clemson University’s Biomedical Engineering Department, to create 3D models of arteries using Simvascular computer programming based off of clinical data from his patients.

As I understand, this group wants to build foundation models using EEG data by pertaining. I’d like to offer my skills in working with machine learning but also clinical data to perhaps with cutting down training time via introducing diffusion. This allows us to potentially cut pre-training completely out of the equation - as it has been done with protein language foundational models successfully. Please refer to DiffDock as an example. We could also tap into clusters from OpenAI’s research fund or San Francisco Compute Company.

I have an entire year ahead of me where I plan to also hopefully publish a paper on the heuristics of these models applied over clinical data and whether little change to training inputs are necessary via diffusion. I have an extensive background in python and recently picked up PyTorch while working on recent Scroll Prize Challenge (https://scrollprize.org/).

Contributor:
Ayan Ansari

• Email: ayan2008627@gmail.com
• Potential Mentors:
  • Mahmoud Zeydabadinezhad (@zeydabadi)
  • Babak Mahmoud
    CV: ayanansari.2023-3.pdf

[kernel-loophole]

@alia-abbas I'm particularly intrigued by the idea of incorporating diffusion to potentially reduce training time and eliminate the need for pre-training, drawing inspiration from successful implementations in protein language foundational models, such as DiffDock. I also wanted to work on this .

[alia-abbas]

Do you know when or how we begin?

…

On Thu, Apr 18, 2024 at 3:15 AM Zulqarnain ***@***.***> wrote: @alia-abbas <https://github.com/alia-abbas> I'm particularly intrigued by the idea of incorporating diffusion to potentially reduce training time and eliminate the need for pre-training, drawing inspiration from successful implementations in protein language foundational models, such as DiffDock. I also wanted to work on this . — Reply to this email directly, view it on GitHub <#50 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APTOUCDKWQ7YPVT77NCHD4DY56MKHAVCNFSM6AAAAABFUOYOSCVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TCNJTGI3TG> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[kernel-loophole]

sorry about the late response .now i am available and looking forward to discuss more .