AIQuS : Artifical Intelligence and Quantum Systems
Research project selected under the 2018 call for proposals
Principal Investigator :
Sylvain Capponi ()
LPT (Theory)
Type of project : Disruptive Project
Date (start/end) : 2018 - 2020
Example of neural networks with a visible layer at the bottom and a hidden one at the top. On the right, it represents a restricted Boltzmann machine with a special connectivity, that can represent efficiently some wavefunctions.
Abstract
Many-body quantum theory appears in lots of physical situations and still represents a major challenge in numerical simulation due to its intrinsic complexity and the size of the objects such as the N-body wavefunction which has an exponential number of coefficients. Therefore, it fits into the "big data science" framework. There are exact results showing that the ground-state of a local Hamiltonian can be represented as a Boltzmann-like neural network with some depth. Similarly, tensor network ansatz, such as the ones used in several algorithms at LPT, also have such a representation. In the field of artificial intelligence, one also needs to apply some learning algorithm to train the network and fix its parameters: this is called "machine learning".
All these concepts have been recently applied to condensed matter physics as well. Our purpose is to improve our knowledge in this rapidly evolving field. Our local expertise in numerical simulation is a strong asset. Moreover, ideas from statistical mechanics or quantum information may prove to be fruitful to justify some heuristic approaches in artificial intelligence. For instance, in order to go beyond standard learning algorithms, it has been shown that it could be more efficient to use DMRG-like algorithm, which is a state-of-the-art numerical technique to tackle one-dimensional quantum systems, or tensor network techniques.
Our project aims at developing national and international collaborations in this field. On the one hand, we will fund travel exchanges for students and young researchers at LPT; on the other hand, we will organize a school on this topic, open to the whole Next community and beyond, with lectures and tutorials provided by world-leading experts in artificial intelligence for quantum systems.