stage: Quantum Machine Learning for Optimal Control Problems H/F

TYPE DE CONTRAT : Stage NIVEAU DE FORMATION : BAC +4 / BAC +5 SPÉCIALITÉ(S) : Expertise / Recherche PAYS / RÉGION : France / Ile-de-France DÉPARTEMENT : Essonne (91) VILLE : palaiseau EDF est labellisé Happy Trainees DESCRIPTION DE L'OFFRE An optimal control problem aims at determining the optimal values of a control variable over a given time horizon in order to minimize a cost function, subject to some stochasticity. One of the many applications of interest for a company as EDF is the computation of optimal portfolio positions (financial or energy production assets) in order to minimize the associated nancial risk (e.g. as in the Black Scholes model). This type of problem was traditionally solved considering model, but solutions based on classical machine learning and deep learning have been studied and have shown many advantages (large size possible, possibility to consider liquidity constraints and transaction costs,...). In parallel, quantum machine learning has emerged as a promising eld, and is developing fast, with newly proposed algorithms as HHL based on quantum circuits and adapted software. The "gate" or circuit model is the standard language to design an algorithm that may be interpreted by a quantum computer. Although a medium scale universal quantum computer may not be available before several years, IBM and Google have already made available quantum circuits libraries (respectively called Qiskit and Cirq) and Google has proposed a quantum extension of its machine learning library, called Tensor flow Quantum. The latter enables the user to quickly design a quantum neural network, based on quantum circuits, train the network (either by simulating a quantum machine or interfacing a real one) and use it. The objective of the internship is to implement and assess the relevance and eciency of state-of-art quantum machine learning algorithms and methods for solving optimal control problems (e.g. hedging problems). The intern will first be able to run quantum simulations (e.g. from Cirq) from a classical computer or a cluster but, in a second step, we could provide quantum computing time on existing machines that are commercially available. PROFIL SOUHAITÉ First experiences with quantum computing, quantum circuits (theory and use of libraries) and/or machine learning will be appreciated.