Lenka Zdeborová honoured by the 2021 Josiah Willard Gibbs Lectureship
Lenka Zdeborová, who recently joined EPFL as Associate Professor of Physics, Computer Science and Communication Systems in the Statistical Physics of Computation Laboratory – part of the School of Basic Sciences (SB) and School of Computer and Communication Sciences (IC) – has always loved mathematics and physics. And this was even before, as a child, she started winning competitions and the physics and mathematics Olympiads in her native Czech Republic.
Zdeborová says at first, like many physics students, she was fascinated by topics like astrophysics and quantum mechanics but during her Masters studies she realized that condensed matter theory and, in particular, statistical physics are really intriguing. One paper in particular led her to a PhD on the interface between statistical physics and computational complexity theory.
“The paper that I was given was basically looking at a computational problem as a statistical physics model, as if the bits were molecules and the rules of the computational problem were the interactions between the molecules and I found it absolutely fascinating. It’s really the broad idea of looking at computational problems and algorithms as objects of physics in the same way that in physics we study black holes and want to understand what they are and how they behave, and this field has stayed with me since.”
Zdeborová takes the computational problems, the algorithms and their behavior as objects that exist and aims to understand why they behave the way they do, with the ultimate goal of improving them and one day understanding what is computationally possible and what is not.
And this is what inspired her presentation this year to the American Mathematical Society in the 2021 Josiah Willard Gibbs Lecture, in which she described the field of physics to study and understand computational and algorithmic problems.
“I’m inspired by the fact that Gibbs is one of the fathers of statistical mechanics who explained the laws of thermodynamics using more fundamental principles and right now in my field we are asking which machine learning problems are actually tractable with computers and which are not. We really don’t have something that could be called the well-established theory of learning, we’re still building this and I would like to think of the sub-field in which I’m working – the applications of statistical physics to computational problems – as contributing to this theory that will hopefully come one day soon.”
Zdeborová learned about the invitation to present the lecture almost a year ago and admits that it was a bit overwhelming, “a distinguished list of people, including Albert Einstein, has come before me. I’m very proud and it was even a bit intimidating although it is a wonderful opportunity to let the world know about my field and about the collective efforts and achievements of myself and my colleagues within it.”
At EPFL Zdeborová is passionate about advancing the theory of what is computable and what’s possible with machine learning and artificial intelligence, as well as fostering greater collaboration with colleagues across laboratories, in different fields and with different ideas. She is already involved in establishing a working group on what she calls scientific machine learning, which she explains is the role that machine learning is playing in changing the very scientific method.
“In sciences we want to understand the objects we study better, the objective is not fixed. We need to come up with the objective so that the machine learning system is useful in the scientific endeavor. It’s a fascinating field that has emerged as machine learning has become very successful in the past decade, it touches many labs and professors at EPFL and it’s one field that I want to help structure, making it visible to the outside world.”