@JohnBostanci @ccanonne_ to follow-up on John's comment, what drew me to this problem initially was this observation that the Harrow–⁠Montanaro testing algorithm is so elegant, but making this tolerant leads to all sorts of issues. turns out, there's a bit of genuine difficulty there!

RT @sitanch: Excited to announce @JordanCotler, @RobertHuangHY, @jerryzli, and I are organizing a workshop at FOCS on quantum learning ⚛️!…

thanks so much to @benbenbrubaker for covering the work of Allen Liu, @AineshBakshi, Ankur Moitra, and me on death of entanglement at high temperature! the result is quite surprising imo and i'm grateful that ben wanted to share the story of our surprise to a broader audience.

This gives you a lot: bc we only need constant error, we can linearize e^(-iHt) ≈ I - iHt, so structure learning falls out from classical shadows-y stuff. We even developed an FPT algorithm for it via a Pauli version of Goldreich–Levin. Check it out! 📄:

If you're interested in a jargon-less explanation of "Learning quantum Hamiltonians at any temperature in polynomial time": @science_eye wrote a great piece on it. Thanks, Lakshmi!

@letonyo @AineshBakshi maybe a reincarnation? we will see if it sticks :)

@roydanroy @AineshBakshi good question! maybe not if we want systems with quantum phenomena. but you could ask to extend classical results to unentangled systems, like how to generalize Glauber dynamics, or prove certain correlation decay type statements. this is pretty open and interesting, i think

The proof admits a preparation algorithm when combined with a new quantum sampling-to-counting reduction. We think pinning down the exact critical temperatures for separability and preparability of Gibbs states is an exciting direction for future work! 📄:

RT @AineshBakshi: Update: Allen Liu won the Best Student Paper award at #QIP2024 for this work and will be giving the talk tomorrow (Thursd…

RT @AviadRubinstein: After all this hard work... finally presenting your brilliant results at FOCS, but only 12 people are in the audience?…

@Qottmann The behavior of p, q in the graph is as desired! p, q being smaller on the negative tail is precisely what we mean by it being flat, though the approximation quality is worse on the rest of the function. Thanks for the questions; feel free to email/DM for follow-ups.