Darshil Doshi @ICML2025 (@darshilhdoshi1) 's Twitter Profile
Darshil Doshi @ICML2025

@darshilhdoshi1

Johns Hopkins | UMD | Brown | IITGN โ€”โ€” ML | Physics

ID: 780333664701861888

linkhttp://d-doshi.github.io calendar_today26-09-2016 09:10:18

24 Tweet

48 Followers

102 Following

IAIFI (@iaifi_news) 's Twitter Profile Photo

Thank you to all who attended the #iaifisummerworkshop! We had an excellent week of interdisciplinary interactions with researchers at all career stages. Genuinely inspiring! Until next yearโ€ฆ โฆAIVO -AI Institutes Virtual Organizationโฉ

Condensed Matter Theory Center (@condensed_the) 's Twitter Profile Photo

NeurIPS2023 NeurIPS Conference New Orleans will have two presentations by CMTC from Maissam Barkeshli and Andrey Gromov graduate students Darshil, Dayal, Tianyu-- this is remarkable for a theoretical physics center given the selectivity of this CS conference CMTC is now an AI/ML center too!

Condensed Matter Theory Center (@condensed_the) 's Twitter Profile Photo

As CMTC is getting more and more involved in AI/ML/DL research, with Barkeshli/Gromov and their students just presenting papers in NeurIPS 2023, it is perhaps fitting that this article was just chosen as a Highlight of the Year in the APS Physics Magazine physics.aps.org/articles/v16/1โ€ฆ

Andrey Gromov (@andr3ygr) 's Twitter Profile Photo

Awesome collaboration with my students Tianyu He Darshil Doshi @ICML2025 Aritra Das @ Neurips! We got a GPT-style model to learn modular arithmetic in-context. There are phase transitions as a function of data diversity, number of in-context examples & visually stunning representations.

Darshil Doshi @ICML2025 (@darshilhdoshi1) 's Twitter Profile Photo

I'm genuinely excited to share our recent work on in-context learning modular arithmetic tasks -- featuring the coolest PCA figure you will see today :)

Irina Rish (@irinarish) 's Twitter Profile Photo

Can one achieve SOTA LLM performance at a much lower bitsize (=>memory/inference costs) than current (post-training) quantization? YES! - by training ternary LLMs - a "sweet spot" between underperforming binary and costly ful-precision ones. Happy to announce our recently

Surya Ganguli (@suryaganguli) 's Twitter Profile Photo

Congrats to John Hopfield and Geoffrey Hinton! well deserved recognition that some important foundations of AI rest on physics! Physics departments take note: understanding and improving AI systems is a new frontier topic for physics, just as biophysics was earlier. Time to hire!

Surya Ganguli (@suryaganguli) 's Twitter Profile Photo

(((ู„()(ู„() 'yoav))))๐Ÿ‘พ Ofir Lindenbaum Physics departments are already hiring in the intersection of physics of ML and ML for physics which needs deep knowledge of both. I am not sure why we are arguing what is physics and what is not. The intersections of the two fields is exciting! The Nobel prize is well deserved.

Darshil Doshi @ICML2025 (@darshilhdoshi1) 's Twitter Profile Photo

๐Ÿšจ ICML 2025 ๐Ÿšจ I'll be at ICML Conference Mon-Fri. DM if you'd like to chat!โ˜•๏ธ Also come check out our poster on cracking Pseudo-Random Number Generators with Transformers! ๐Ÿ•š Tuesday @ 11am #โƒฃ E-1206 ๐Ÿ”—arxiv.org/abs/2502.10390

๐Ÿšจ ICML 2025 ๐Ÿšจ

I'll be at <a href="/icmlconf/">ICML Conference</a> Mon-Fri. DM if you'd like to chat!โ˜•๏ธ

Also come check out our poster on cracking Pseudo-Random Number Generators with Transformers!
๐Ÿ•š Tuesday @ 11am
#โƒฃ E-1206

๐Ÿ”—arxiv.org/abs/2502.10390
Ekaterina Lobacheva (@katelobacheva) 's Twitter Profile Photo

Our new paper got #ACL2025 oral! ๐ŸŽ‰ If you're interested in LLM training dynamics, its phases, and how scaling affects them โ€” check it out! Mila - Institut quรฉbรฉcois d'IA x.com/mirandrom/statโ€ฆ