Quantum Machine Learning: The Next Frontier with Iordanis Kerenidis

Automatic TRANSCRIPT

All right everyone I am on the line with Max Welling Max is Vice President of technologies. At qualcomm, Technologies Netherlands and a full professor at the University of Amsterdam. Max was a guest of the podcasts just about a little over a year ago in May of last year where we talked about gauge Equa variant CNN's generative models and the future of ai and I am super excited to welcome you back to the show. To catch up on what you've been up to and dig into some really interesting topics. Hi, Sam. Thank you for having me again on the show absolutely I I'm really looking forward to our conversation You may know this, but last year's podcast that that we did was the second most popular show of the year. So does the the second most downloaded show for the the podcasts and I am sure this one will be just as hot So looking forward into it before we jump into our topics which include topics like neuro. Federated Learning graph they're all networks. Give us an update on what you've been up to at qualcomm and the you know the current focus of your research at the university. Yeah. So I think many of the topics that we talked about last year are still You know hotly pursued at the university and at Qualcomm So we're doing a lot of generative modelling. where we look at, you know, how can we build models that can generate high dimensional data like images and audio in a very realistic way trained completely unsupervised from from data we are also looking at still at in variances and said of symmetry in Dade, in how we can embed those into these models. and in particular, also looking at graph neural networks with applications now, also to dry molecules. So we've started the project where we try to predict the properties of molecules and can be very naturally model with graph neural net because you know molecule is a little bit like a graph and we've made these models also symmetric two rotations in in Three D. space if I take a molecule and I rotated around, then sit of the internal representations of that narrow network need to transform in a coherent way under these rotations and since it's the graph near on that, it's also it's also symmetric under permutations of the nose. and so all of these symmetry is are built into the neural network and being trying to predict properties for drugs, for instance, and in that in the context of you know if we can find something that is good against a go of it. but we've also looked at efficient learning Trent training networks. So very limited precision is mostly worked as being done at qualcomm because of crispy wanted to run our phones an innocent of lowest the lowest amount of energy and so to to have the highest accuracy but using the least amount of power possible. and so we've been a lot of research and trying to make these. sort of complex neural network computations, very low precision, and as an additional sit of new tracking their revolt have been looking at quantum computation now, and in multiple ways, the first of all quantum chips might be around a corner can we train get can be designed good neural networks that would. Run on quantum chips but also just look at the mathematics off these of of quantum mechanics and see if that quantum of the mathematics is is sort of a new language even for classical neural networks. So yeah, there's there's a lot of things happening and then and then there's of course it neural augmentation that we will be talking about We have been applying these ideas to different application areas in Martyrs Communication also at the university, but also on. Princeton's to to my Mo- detection and also to era correction decoding and also to to channel estimation. Awesome. Awesome. Interesting to hear you bring up quantum and quantum machine learning You know that's something. We can maybe spend a few minutes on a curious on your your overall take I've been trying to myself to get a sense for whether I. Think it's worthy of all the hype it's getting in how far away it is and how relevant it is machine learning and they I what what's your take on it? Yeah. They could easily spend an hour on this. I'm very, very excited about this stuff. So. So for me at this point it is mostly intellectually interesting and I have a physics background. So my my supervisor of US GARAGE TOFU IS A. Nobel, prize onset sort of standard model quantum, a standard model quantum mechanics, and he has a very particular and interesting view on quantum mechanics, which is very controversial. I would say but and and and if you is that, in fact there is A. World there, which is not logical where the world is just in classical state at the very, very small small was level. We don't know what that state is or what the e even how to describe it, and so the mathematics to describe that sort of. Classical Senator Automata is happens to be quantum mechanics. A quantum mechanics is a mathematical tool like complex analysis is a mathematical to to solve. Partial differential equations, ordinary different differential

Coming up next