We connect physics and computer science to develop secure and scalable algorithms and simulation methods, classical and quantum.

Algorithms for safe AI, powered by physics.
We use current developments in theoretical physics – from statistical mechanics of disordered systems to quantum many-body theory – to build algorithms for complex computational problems. By operationalizing physical principles, we reap efficiency benefits for a variety of real-world applications relevant to AI, including combinatorial optimization, machine learning, and multilinear algebra. We focus especially on computation under constraints, a computational approach to building safe AI systems that respect values and regulations. Although our research is fundamental, we strive to deliver solutions that are ready to deploy on the best suited computing hardware, be it quantum processors, GPU servers, or laptops.

Theory of quantum many-body phenomena
In our lab, algorithm design begins with understanding – and sometimes discovering – the physics that governs the computing substrate. This enables us to build protocols that are native to quantum architectures, removing as many latency-inducing layers as possible between the hardware and the problem at hand. To ensure this, we delve into the physics of complex many-body systems out of equilibrium using a combination of analytical techniques and numerical methods, as well as simulation on quantum hardware. In the process, we develop open-source code libraries for the simulation of quantum systems (cotengra, MatchCake, mdopt).

Encoded and encrypted quantum computation.
To unlock the full potential of quantum computation, we must counteract the inherent tendency of quantum systems to decohere. Error correction with surface codes and concatenation provides a path to fault tolerant quantum computing, but qubit overhead poses a major obstacle to scaling. Good quantum error correcting codes with high encoding rates offer a shortcut to fault tolerance, provided we can find and efficiently deploy them in practice. We are developing principled and hardware-aware approaches to code discovery. In tandem, we are building powerful decoders based on tensor networks for accelerated GPU decoding, gearing towards real-time error correction. Moreover, we are devising algorithmic primitives for encrypted quantum computation that protect the privacy of sensitive information when quantum computation is accessed as a cloud service.
