iiinova — where your digital self keeps evolving

Filippo Vicentini

@philipvinc.bsky.social

Professor of Quantum Physics and AI at Ecole Polytechnique in Paris. Core contributor to the Neural Quantum State simulation methods, and to open science efforts.

Posts Replies Media

Filippo Vicentini @philipvinc.bsky.social · 20/07/2025

Today I learned that "Antisimmetric Neural Networks" are patented. @gppcarleo.bsky.social and several other researchers in #VMC #NQS should be careful about lawsuits from Google...

patentimages.storage.googleapis.com

030

Filippo Vicentini @philipvinc.bsky.social · 09/07/2025

000

Filippo Vicentini @philipvinc.bsky.social · 09/07/2025

Credit goes to Antoine who pushed this through quickly. This was also inspired by years-old discussions with Riccardo Rossi on sampling gradients (which we tried hard, but we believe we can match with this at a lower cost) and Ao Chen who mysteriously mentioned it in some paper before removing it.

130

Filippo Vicentini @philipvinc.bsky.social · 09/07/2025

For 'simple' problems sampling α=2 seems good, but for harder ones you benefit from an over dispersed distribution. Worry not! This is FREE and you can even autotune α so VMC is not harder. You can find a #NetKet based repository, fully reproducible github.com/NeuralQXLab/...

110

Filippo Vicentini @philipvinc.bsky.social · 09/07/2025

We wondered "Are we sampling right?" And the answer seems "probably no". When doing #VMC on hard #quantum problems, |ψ(x)|² is optimal to get good "Energy estimates" but you need precise gradients to converge! Turns out, you converge better by looking at |ψ(x)|ᵅ . arxiv.org/abs/2507.05352

150

Filippo Vicentini @philipvinc.bsky.social · 02/04/2025

4251

Filippo Vicentini @philipvinc.bsky.social · 25/02/2025

Let me conclude that this will be the starting point of a wider effort to package and cleanup NQS dynamics simulation methods. We'll soon start a new effort to collect various codes for dynamics NQS methods and ensure that they can be pitted one against the other!

020

Filippo Vicentini @philipvinc.bsky.social · 25/02/2025

I'd like to congratulate Luca, that 'leads by example' by being open, transparent and ensuring total reproducibility of his results. I can only encourage all other people in our field (and beyond) to do the same!

120

Filippo Vicentini @philipvinc.bsky.social · 25/02/2025

We also add all the states we obtained through our simulations, and you can compute quantities on them yourself. Just `nqxpack.load("data/states/t=XYZ.nk` and compare to our data! The training scripts to train a 6x6 or 10x10 ViT or ConvNet are included, as well as an usable, well documented code.

120

Filippo Vicentini @philipvinc.bsky.social · 25/02/2025

Do you want to simulate large spin systems with NQS? Yourself? Automatically? We have just released the accompanying code for the Systematic study of projected neural dynamics, together with Luca Gravina and Vincenzo Savona! Check it out! github.com/NeuralQXLab/...

160