The NEASQC project is organising a series of webinars that are not a simple overview of the NISQ Quantum Computing use cases investigated by our project, but a platform to learn from and exchange with the best experts in the fields covered by NEASQC. Our Work Package 5 prepared a programme of four weekly webinars from 5 to 24 November, dedicated to Machine learning and optimization methods. Each week, a distinguished guest speaker will share their insights on the week’s topic, and a NEASQC expert will then explain how NEASQC is addressing the topic through a specific use case.
On 5 November 2021 from 17:30 to 19:00 CET, we will address Quantum Machine Learning
Agenda
- 17:30 – 18:10 Guest speaker: Dr Maria Schuld, Senior Researcher and Software Developer at Xanadu
Perspectives of quantum machine learning for real-world applications - 18:10 – 18:15 Technical questions
- 18:15 – 18:35 Reinforcement learning for optimal stock management (by Reda Alami, TotalEnergies)
- 18:35 – 19: 00 Open discussion and expert Q&A session
Moderator
Dr Vedran Dunjko, Leader of the Work Package 5 (Machine Learning & Optimisation) in the NEASQC project, Assistant Professor at Leiden University
Speakers
Dr Maria Schuld, Senior Researcher and Software Developer at Xanadu
Perspectives of quantum machine learning for real-world applications
Abstract: Machine learning is often mentioned as one of the most promising applications for early-stage quantum computing. This is not self-evident, because problems in machine learning are the opposite of what quantum computers are good at: they take large inputs, have unstructured solutions, are mathematically complex to model, and they are already solved astonishingly well by classical computers. Why, then, is quantum machine learning such an exciting field? This talk will present ways in which quantum computers slot into machine learning theory surprisingly elegantly, for example by using them either as probabilistic models or as distance measures (also known as “kernel methods”), which can be trained using existing deep learning tools. Even though a lot still has to be done to arrive at “real-world applications”, from a scientific perspective these points of connection give us rich frameworks to start exploring the use of quantum computing for data mining.
Reda Alami, Research Scientist – Reinforcement Learning at TotalEnergies
Reinforcement learning for optimal stock management
Abstract: The inventory stock control is one of the most significant problems in the supply chain management process of a firm. Reducing its stock costs helps gaining in performance and competitiveness. The inventory stock management includes aspects such as controlling and overseeing purchases from suppliers as well as customers, maintaining the storage of stock, controlling the amount of product for sale, and order fulfillment. A decision maker (learning agent) observes the random stochastic demands and local information of inventory such as inventory levels as its inputs to make decisions about the next ordering values as its actions. Since the inventory on-hand (the available amount of stock in inventory) unmet demands (backorders), and the existence of ordering are costly, the optimization problem is designed to minimize the overall cumulative costs. As a result, the objective function is to reduce the long-run cost (cumulative reward) whose components are linear holding, linear penalties, and fixed ordering costs. In most inventory management policies, this is done using basic heuristics that are not always able to account for the complexity of the system and the stochasticity of the demand. This results in two possible scenarios: the first is to exceedingly order which results in paying unnecessary costs, the second is to make an insufficient order which results in unsatisfied demands. In order to minimize inventory management costs, we propose a deep reinforcement learning based approach. Indeed, stock management can be modeled as a sequential decision-making process under uncertainty which is often written as a Markov decision process (MDP). In this case, reinforcement learning provides robust solutions for this kind of tasks. Experimentally, the performance of the proposed method is found to be satisfactory in comparison with the main state of art proposed solution namely, the (s,Q) policy.
Events 
News 
