Speaker
Description
Tensor Networks (TN), originally developed in the context of quantum many-body physics, have recently emerged as powerful and interpretable Machine Learning (ML) architectures. The objective of this talk is twofold. Firstly, provide an intuitive and practical perspective on TNML methods, starting from the seminal work that initially established its connection to supervised learning. Secondly, present QChaiTEA, the ML application of the QTEA framework, describing its design and structure. This library provides comprehensive coverage of the entire training and analysis process for a quantum-inspired ML model. It encompasses the embedding map from classical data to tensor network states, the modelling through different TN ansätze, the optimisation step, and the extraction of meaningful quantities to describe and interpret the trained model. The library has been designed to be highly flexible. It provides a variety of embeddings, including the conventional feature map in separable states and MERA convolutions. Furthermore, it implements two optimisation procedures: one being backpropagation and the other one inspired to the DMRG algorithm. Finally, an application in High Energy Physics is showcased. In this context, we compare TNML against classical ML in the jet tagging task for fast inference at the trigger level. We show that unique characteristics of TNs, such as computational efficiency and interpretability, play a crucial role.
| Sessions | Quantum Machine Learning: |
|---|---|
| Invited | No |