 Okay, so I'm going to describe a new obfuscation scheme with my co-author Craig Gentry. So this obfuscation scheme uses tensor products, so the title obfuscation using tensor products. So it's, first of all, it's a purely algebraic construction. And over the last few years, we have seen some obfuscation candidates. They were based on multilinear maps and multilinear maps themselves based on noisy encodings. So our scheme is completely different in that regard. We do not use any noisy encodings. It's purely algebraic, as I said. And it uses matrix algebra and some tensor algebra, tensor symmetresses. It still uses Barrington's theorem, which is what the multilinear map-based schemes were using. So this also works for, so far it works for branching programs and so also for NC1 circuits. Okay? So the main idea is that we take a branching program and normally you could try doing a chelonization of all the matrices in your branching matrix product program. So that obviously is not secure. So what we really have to do is build these chelon matrices in a dynamic fashion. Dynamic meaning that the chelon matrix will actually depend on the input on which you're trying to evaluate the program. Okay? And these chelon matrices for each input will yield a completely new random chelon matrix. But obviously we have to give you some gadgets to be able to implement this strategy. So before I describe something more about it, when he was here he would probably be already shocked that I'm claiming some efficiency things about our scheme, but luckily he's not around, I haven't seen him. So this thing is also much more efficient than the multilinear map-based scheme. So we ran some back-of-the-end envelope calculations. So if you have a branching program with a million steps and the input is of 100 bits, your obfuscated program will fit in 256 gigabyte, right? Which is nowadays a server will have that kind of RAM, at least. And to evaluate it on a single core Intel i7, I guess, it will just take 1 to 10 minutes. So which, like I said, comparatively speaking, it's fast. So I don't want to cite the numbers for the multilinear map-based schemes. I did talk to Shy about it and he was like, okay, let's not talk about it. Okay, so what about security of the scheme? So this, since we have this algebraic construction and we have matrices, you know, each of those things you can think of as variables which are hidden and are all secret. So you can build a whole bunch of polynomial equations and you can try to solve them. So this will fall into the realm of multivariate cryptography. And as we know, over the last couple of decades, people have worked on these things and there were a lot of attacks also. So in this one, we actually prove a theorem that given all the information you get from the obfuscated program, whatever polynomial equations you can build, generic Grobner basis techniques are going to be inefficient. So this would include the Kipnis-Shamir real linearization techniques and its generalization which is called the Excel methodology by Kottua, Shamir, Patarin, and I think there was another author there. So we prove that such generic Grobner basis techniques will require exponential number of monomials so that would require exponential attack. And we also give arguments as to why algebraic techniques in general will not succeed. But obviously we can't give a lower bound because that will be equivalent to proving circuit lower bounds or something similar to that. So one takeaway from this, because these are new assumptions, is that I can or we can formulate this problem which you can think about. We call the one more masked tensor problem. So here you are given lots of tensor products masked by a single matrix and you are supposed to come up with a new masked tensor problem which is not a scalar multiple of the previous one. So we consider this problem to be hard and our security is related to this problem. Thank you.