Oleksii Malyshev @malyshevjr.bsky.social · 10/04/2026Thank you very much for putting out such a coherent vision; lately, I was also getting more and more convinced that somewhat paradoxically having agents at our disposal shouldn’t change *anything* about the way we train people at universities, either at undergrad or higher levels 010 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026Huge thanks to Ivan Morera, Simon Linsel, Fabian Grusdt, Annabelle Bohrdt @abohrdt.bsky.social and Eugene Demler! @dulwichquantum.bsky.social I hope we could make it to the meme-driven feed of quant-ph, thank you very much in advance! 011 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026It was a really interesting project to work on, and I believe that distance learning will find applications beyond mapping equilibrium phase diagrams: dynamical phase transitions, reconstructing Berry curvature, switching from the Fisher to Wasserstein geometry—sky is the limit! 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026We applied our method to a what seems gazillion of systems, and it... worked! We reconstructed phase diagrams for symmetry-broken and topological, spin and fermionic systems, and even saw a good critical scaling for the Fisher information reconstructred directly from snapshots! 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026There's more to the reconstructed divergences: for the neighbouring states they are an approximation to the Fisher information (FI). This allows us to probe the nature of the discovered phase transitions too: FI is related to the fidelity susceptibility, characterised by its own critical exponent 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026As a result, we reconstruct the full matrix of pairwise distances between the states, and in many cases it is easy to see the phase structure already there. But, of course, eyeballing is not a way forward, so we run HDBSCAN clustering on it and read out phases from the cluster! 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026Learning the ratio r(x) = q(x)/p(x) is much easier: we just train a neural classifier which discriminates between p and q given a snapshots x, and read out r(x) from the ratio of *posterior* probabilities (assuming our classifier is Bayes optimal!) 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026One might naïvely calculate these divergences by approximating full normalised probabilities p(x) for each batch of snapshots. This is costly and hard! Thankfully, there's a way around: one actually needs only *ratios* of such probabilities to calculate any of f-divergences! 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026First, we argue that the natural distances between clouds of snapshots are *statistical* distances. These include KL, Jensen-Shannon, Hellinger divergences—and many more going under the umbrella of *f-divergences*. We want to calculate any of these for all pairs of snapshot clouds 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026However, to cluster a bunch of points in some vector space you don't really need the points themselves. In fact, pairwise *distances* between these points are already sufficient to do clustering. And this was exactly how we unleash the Occam's chainsaw 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026So far, people mostly focused on what we call *feature learning*: every bunch of snapshots was compressed into something low-dimensional (think PCA, diffusion maps, etc.). Then, one would try to make sense out of these representations (e.g. cluster them) 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026Assume experimentalists prepared some fancy quantum states, measured them in the computational basis, and handled all these data to us. They say "Do something machine learning-y, tell us if there are phase transitions or stuff like that". How do we proceed? 101 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/03/2026🚨Paper alert🚨 We propose a new way to analyse quantum simulator snapshots: 📏 distance learning 📏 It maps symmetry-broken and topological phases diagrams, probes criticality, universality, and many-body physics more generally 🧵 scirate.com/arxiv/2603.1... 131 3
Oleksii Malyshev @malyshevjr.bsky.social · 12/11/2025From my perspective, switching to standard time actually gets us an important hour of daylight in the morning; I don't care so much about the evenings, it's gonna be dark anyway, and I am at work regardless. Instead, light in the mornings really helps to wake up and not feel groggy until the lunch. 120 2
Oleksii Malyshev @malyshevjr.bsky.social · 03/07/2025Just heard „The World’s Worst Girlfriend“ on Polish Radio Czworka, these guys keep up! 010 1
Oleksii Malyshev @malyshevjr.bsky.social · 06/02/2025Papa Don’t Preach Holiday Into the Groove Who’s That Girl The Look of Love 020 2
Oleksii Malyshev @malyshevjr.bsky.social · 03/02/2025I think I suffered from one during my teenage years (my shtick was making sure that almost *every* action is performed (3n + 1)times), but then it thankfully subsided! 010 1
Oleksii Malyshev @malyshevjr.bsky.social · 30/01/2025This looks very cool! Do you possibly have an idea how difficult would it be to implement this in Jax? 110 1
Oleksii Malyshev @malyshevjr.bsky.social · 23/01/2025Maybe smart people came up with scripts for this long time ago, but I feel like the freshly announced OpenAI Operator will be a great tool to check fresh arxiv submissions for papers on a specific topic… 020 2
Reposted by Oleksii MalyshevFlorian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 22/01/2025In March 2017, there were 'Marches for Science' all around the world. Is anything similar planned now (it appears even more urgent)? 074 7
Oleksii Malyshev @malyshevjr.bsky.social · 21/01/2025Hey guys, I would just like to say thank everyone on Bluesky who made me realise I wasn't the only one stressing about Elon Trump's (yesterday's?) behaviour. Of course my subscriptions might not be representative of the outer world, but hey, if it keeps me more sane, that's good enough for now 020 2
Oleksii Malyshev @malyshevjr.bsky.social · 21/01/2025Absolutely the same sentiment, I am really concerned about the apparent normalisation of the "rule of power" and the readiness of media to adopt to the emerging narrative rather than shape it 010 1
Oleksii Malyshev @malyshevjr.bsky.social · 20/01/2025For people outside of the field, could you please provide some context, sounds super intriguing :) 110 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/01/2025I hope Mr Musk who really likes making decisions based on public opinion polls will take this one seriously www.theguardian.com/technology/2...theguardian.comElon Musk knows little about UK and Germany and should not interfere – pollExclusive: Majority polled by YouGov viewed billionaire negatively – except among Reform UK and AfD voters 010 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/01/2025Btw, I find it kind of funny how I started my PhD just after you released the second album, and how it took me until your third album release to defend (which is much longer than expected, the whole academic journey should have taken ~3 years, but ah well, we know how crazy the world went in 2020) 010 1
Oleksii Malyshev @malyshevjr.bsky.social · 17/01/2025I am already warming up by re-listening the first two albums!! 110 1
Oleksii Malyshev @malyshevjr.bsky.social · 12/12/2024I can spot a verrrry familiar LaTeX template:) 000 0
Oleksii Malyshev @malyshevjr.bsky.social · 09/12/2024This is uncanny how just today I was thinking if there's a better way to approximate A=B B^T without directly evaluating the matrix product, and then I see this in my feed... In my case B is of dimension r x N, r << N and the standard number of operations O(r^2 N) is too costly 010 1
Oleksii Malyshev @malyshevjr.bsky.social · 03/12/2024Does it have “revtex looking” templates? Otherwise looks very cool!! 100 1
Oleksii Malyshev @malyshevjr.bsky.social · 01/12/2024I once read a The Guardian editorial by Stewart Lee where mocked Elon by saying something along the lines “… Twitter (currently known as X)”, I thought that was funny:) 020 2
Oleksii Malyshev @malyshevjr.bsky.social · 30/11/2024Ah, gee, that’s what the OP has too to a much larger degree of perfection… Very impressive! 120 2
Oleksii Malyshev @malyshevjr.bsky.social · 30/11/2024I’ll put my two cents: I had a Greek housemate doing engineering, and he really had it easy with the Greek keyboard layout: he never had to type anything like \alpha in LaTeX, he was just like: α 120 2
Oleksii Malyshev @malyshevjr.bsky.social · 20/11/2024I can't guarantee its accuracy, but I've heard a similar story that in personal conversations at the conferences Walter Kohn himself was expressing his surprise (if not disbelief) at the fact that people made DFT work :) 010 1