 All right, so it's an announcement for the library that we've just released on Monday, and it's about the fully homomorphic encryption and You all you by now you've been to at least one talk about FHE so you know what it's about and While I'm talking you can just meditate on this picture above That explains it pretty much and if you want to know more about FHE and latest cryptography you can There are still a few if you seconds available to a screenshot or snapshot this these two links one is the collection of Collection of papers about FHE and the second one is a very good book a review FHE and just a few points that I want to emphasize about the algorithm that we use in this in this library is first that There's a very huge suffatex expansion 16,000 times two kilobytes per bit of plain text Second the operations on suffatex all the gates the binary gates that we have on beats are very fast but what takes time is the bootstrapping because all the separations introduce errors and after each operation we need to get rid of them and that's what takes the majority of time and Okay So there are two implementations of the FHE on torus one of them is pure C++ Called FHE and the second one is QFHE which is based on CUDA. So why do we need Python then? Well, we're touch we are targeting GPUs So we can hide all the overheads of Python behind our synchronous Calls to the kernels on GPUs so it doesn't really matter that the Python is slow because the GPU is fast and All this score generation that we can do on Python helps a lot with optimizing for performance and I use that a lot in the library And of course, Python is easy. It's easy to install. It's easy to see what is happening It's easy to test easy to hack to do something with it And that's how an example looks like that's what you can find in the repository. It's just You create a key There's no point to probably here. Yeah, you create a key you encrypt some bits You run a gate on them and you decrypt you decrypt them with a private key back And you can check that you got what you expected in this case. It's an end gate right on the sequence of bits Now for the most important part the performance of course, and that's what what we have in the in the title First I compare it compared with the TPG. Of course, it's a bit, you know Not very fair because TPG uses CPU, but we have a hundred hundred times speed up As compared to the FHG. That's using FFT for bootstrap That's what TPG uses and also we can compare it to QFHE which uses entity. Okay, which is a Fully like transform on finite fields and it has about the same speed, but new FHG supports open sale as well It's written in Python instead of QDA. So, you know, it's still good And if you want to see it you can check it out on this repository here Thank you. Yeah, that's it