Jonathan Shi
jnshi.bsky.social
Optimization/sampling theory for highly non-convex problems from physics. 3 chip tapeouts incl. Intel 3. Won a COVID forecasting contest by 200,000× likelihood.
👤 Joined on 16 November 2024
22
Followers
7
Following
18
Total Posts
📜 Kullanıcı Geçmişi
- 16 Kas 2024, 04:13
🎉 Bluesky'e katıldı - İlk kullanıcı adı:
@jnshi.bsky.social
🚫 Blokladığı Kullanıcılar
❤️ Beğendiği Gönderiler
This was an up-and-down journey full of a lot of learning and discovery. However, the embarrassment of missing the fact that we could’ve gone for the “local” regularity properties straight away on top of the free-probability and cavity interpolation theory is real.
Show Postcc: @jnshi.bsky.social - Because the tag broke in the first post, for some reason.
Show PostSticking to some good-old math: arxiv.org/abs/2609.08169 This came out of the PHA series of papers as @jtnshi & I observed one could Guerra interpolate the planted-SK with SL field to CWRF. We were curious to develop a sharp understanding of the overlaps under the Gibbs at high-temp for the CWRF.
Show Postto design a "tube" around good paths where the TAP solutions are unique, and then reason about these "local" regularity properties by defining a smooth extension to the resolvent that drive the PHD process at all points inside the cube.
Show PostThis requires amending the prior error analysis between the ideal and algorithmic SDEs to only use local regularity over the points (forming a non-conex set) where the TAP Hessian is strongly convex. One can loan Celentano's argument with the "stability" of the SL process
Show Post