Optimal mass transport: a new approach to quantum machine learning

Where: Zoom platform - 2 p.m.

Speaker

• Dr. Giacomo De Palma  - MIT (USA)

Abstract

Quantum computers can provide a potential exponential speedup in machine learning tasks. One of the most promising applications of near-term quantum computers is the quantum version of the Generative Adversarial Networks (GANs). GANs provide an algorithm to learn an unknown target probability distribution with extraordinary capabilities in computer vision and artificial intelligence, such as generating fake photographs or videos that look realistic to the human eye. Quantum GANs, the quantum version of GANs, provide an algorithm to learn an unknown target quantum state, with potential applications in quantum simulations of molecules, drug discovery and algorithmic trading. A crucial factor for the success of GANs is the choice of the cost function that measures the quality of the approximation. The best choice for classical GANs comes from the theory of optimal mass transport, and is provided by the earth mover's distance. I will present a generalization of the earth mover's distance to the quantum states of n qubits, which provides a new approach that makes quantum learning more stable and efficient. The quantum GANs based on this distance are capable of learning a broad class of quantum data, with remarkable improvements over the previous proposals.

Scientific Coordinator

• Dr. Iacopo Carusotto