Search Talks

I will present recent progress in building a rigorous theory to understand how scientists, machines, and future quantum computers could learn models of our quantum universe. The talk will begin with an experimentally feasible procedure for converting a quantum many-body system into a succinct classical description of the system, its classical shadow. Classical shadows can be applied to efficiently predict many properties of interest, including expectation values of local observables and few-body correlation functions. I will then build on the classical shadow formalism to answer two fundamental questions at the intersection of machine learning and quantum physics: Can classical machines learn to solve challenging problems in quantum physics? And can quantum machines learn exponentially faster than classical machines?

Zoom link:  https://pitp.zoom.us/j/97994359596?pwd=UlBwc2hoSkNzWlZvM1o1RWErU1U2QT09

Google's TPUs were exclusively designed to accelerate and scale up machine learning workloads, amid the ongoing planet-wide race to build faster specialized hardware for artificial intelligence. But one must surely be able to use this hardware for other challenging computational tasks, right? We explored how to turn a TPU pod (2048 TPU v3 cores) into a dense linear algebra supercomputer to e.g. multiply two matrices of size 1,000,000 x 1,000,000 in just 2 minutes. We then used this power to perform a number of quantum physics and quantum chemistry computations at scale. For instance, we recently completed two largest-ever computations: a Density Functional Theory DFT computation of electronic structure (with N = 248,000 orbitals), and a Density Matrix Renormalization Group DMRG computation (with bond dimension D = 65,000). Cloud-based TPU pods and GPU pods are accessible to anyone and are poised to revolutionize the scientific supercomputing landscape.

Zoom link:  https://pitp.zoom.us/j/97006251134?pwd=WHhLa2pRUno0LzJjbzVnL2tsdGhOUT09

In this talk I will share our recent work in developing statistical models based on machine learning methods. In particular, I will discuss posterior sampling in low- and high-dimensional spaces and connect this to two ongoing projects: measuring the small-scale distribution of dark matter and estimating the expansion rate of the Universe. I will discuss how the speed and the accuracy gained by these models are essential for the large volumes of data from the next generation sky surveys. I will finish by mentioning a few other projects and a new initiative for interdisciplinary collaboration in astrophysics and data sciences. 

Zoom link:  https://pitp.zoom.us/j/98316228305?pwd=UWwrZkIwUG1QZFBkYzc1eVdNSW1Ldz09

New quantum simulation platforms provide an unprecedented microscopic perspective on the structure of strongly correlated quantum matter. This allows to revisit decade-old problems from a fresh perspective, such as the two-dimensional Fermi-Hubbard model, believed to describe the physics underlying high-temperature superconductivity. In order to fully use the experimental as well as numerical capabilities available today, we need to go beyond conventional observables, such as one- and two-point correlation functions. In this talk, I will give an overview of recent results on the Hubbard model obtained through novel analysis tools: using machine learning techniques to analyze quantum gas microscopy data allows us to take into account all available information and compare different theories on a microscopic level. In particular, we consider Anderson's RVB paradigm to the geometric string theory, which takes the interplay of spin and charge degrees of freedom microscopically into account. The analysis of data from quantum simulation experiments of the doped Fermi-Hubbard model shows a qualitative change in behavior around 20% doping, up to where the geometric string theory captures the experimental data better. This microscopic understanding of the low doping limit has led us to the discovery of a binding mechanism in so-called mixed-dimensional systems, which has enabled the observation of pairing of charge carriers in cold atom experiments.
Intriguingly, mixed-dimensional systems exhibit similar features as the original two-dimensional model, e.g. a stripe phase at low temperatures. At intermediate to high temperatures, we use Hamiltonian reconstruction tools to quantify the frustration in the spin sector induced by the hole motion and find that the spin background is best described by a highly frustrated J1-J2 model.

Zoom link:  https://pitp.zoom.us/j/99449352935?pwd=cXdYYTJ2c1hVZ014SWRwZi9LRjQ3dz09

While quantum computers are naturally well-suited to implementing linear operations, it is less clear how to implement nonlinear operations on quantum computers. However, nonlinear subroutines may prove key to a range of applications of quantum computing from solving nonlinear equations to data processing and quantum machine learning. Here we develop algorithms for implementing nonlinear transformations of input quantum states. Our algorithms are framed around the concept of a weighted state, a mathematical entity describing the output of an operational procedure involving both quantum circuits and classical post-processing.

Zoom Link: https://pitp.zoom.us/j/92831825506?pwd=T2VUQ2M2QlZERmRmUHZ0T1VOelkzZz09

The sign problem is arguably the greatest weakness of the otherwise highly efficient, non-perturbative Monte Carlo simulations. Recently, considerable progress has been made in alleviating the sign problem by deforming the integration contour of the path integral into the complex plane and applying machine learning to find near-optimal alternative contours. This deformation however requires a Jacobian determinant calculation which has a generic computational cost scaling as volume cubed. In this talk I am going to present a new architecture with linear runtime, based on complex-valued affine coupling layers.

This course is designed to introduce modern machine learning techniques for studying classical and quantum many-body problems encountered in condensed matter, quantum information, and related fields of physics. Lectures will focus on introducing machine learning algorithms and discussing how they can be applied to solve problem in statistical physics. Tutorials and homework assignments will concentrate on developing programming skills to study the problems presented in lecture.

This course is designed to introduce modern machine learning techniques for studying classical and quantum many-body problems encountered in condensed matter, quantum information, and related fields of physics. Lectures will focus on introducing machine learning algorithms and discussing how they can be applied to solve problem in statistical physics. Tutorials and homework assignments will concentrate on developing programming skills to study the problems presented in lecture.

This course is designed to introduce modern machine learning techniques for studying classical and quantum many-body problems encountered in condensed matter, quantum information, and related fields of physics. Lectures will focus on introducing machine learning algorithms and discussing how they can be applied to solve problem in statistical physics. Tutorials and homework assignments will concentrate on developing programming skills to study the problems presented in lecture.

This course is designed to introduce modern machine learning techniques for studying classical and quantum many-body problems encountered in condensed matter, quantum information, and related fields of physics. Lectures will focus on introducing machine learning algorithms and discussing how they can be applied to solve problem in statistical physics. Tutorials and homework assignments will concentrate on developing programming skills to study the problems presented in lecture.

The sustained storage, transmission, or processing of quantum information will likely be a non-equilibrium process that requires monitoring the system and applying some form of feedback to produce fault-tolerance.  In this talk, I will discuss a class of models based on random quantum circuits with intermediate measurements that display a similar phenomenology to standard models for fault-tolerance, including the existence of a threshold, but with several helpful simplifications.  However, naïve realizations of the threshold require an exponential number of repetitions of the experiment to fully explore the output space of the intermediate measurements.  Recently, it has been proposed that this problem can be circumvented by developing efficient entanglement “decoders” that have close parallels to quantum error correction decoders.  We show how to leverage modern machine learning tools to devise a neural network decoder to detect the phase transition. We then study the complexity and scalability of this approach and discuss how it can be utilized to detect entanglement phase transitions in generic experiments.

Zoom Link: https://pitp.zoom.us/j/99123641139?pwd=VmkyR3BSNWF5bURVYmFVakp0ZkNRZz09

This course is designed to introduce modern machine learning techniques for studying classical and quantum many-body problems encountered in condensed matter, quantum information, and related fields of physics. Lectures will focus on introducing machine learning algorithms and discussing how they can be applied to solve problem in statistical physics. Tutorials and homework assignments will concentrate on developing programming skills to study the problems presented in lecture.