Seminarium Fizyki Politechniki Wrocławskiej
PWr, bud. A1, sala 322
The physical foundations of machine learning and the future of optical neuromorphic computing
dr inż. Andrzej Opala
Uniwersytet Warszawski, Instytut Fizyki PAN w Warszawie
We have entered a new era of innovations, driven by rapid advances in machine learning (ML). Over the past decade, ML systems such as artificial neural networks have become valuable tool used in industry, research and everyday life. Neural networks excel in pattern recognition, prediction of non-linear processes, and language processing. In scientific research, neural networks are used to control quantum devices, design new materials and analyse experimental data. This year, the Nobel Prize Committee recognized the profound impact of machine learning, highlighting its fundamental connections to statistical physics, nonlinear dynamics, and complex systems. First part of this seminar will focus on the basic principles of ML system formulated by John Hopfield and Geoffrey Hinton, including the concept of Hopfield network, Boltzmann Machines, Feedforward neural network and role of backpropagation algorithm [1].
The current challenges of classical software machine learning include the high energy consumption required to process large amounts of data and the limitations of conventional electronic components. Therefore, the development of novel high-speed, low-power computing systems that go beyond standard architectures is essential for future technological progress. Exciton-polaritons emerge as a promising solution, combining the properties of light and matter in one physical platform, which could overcome the limitations of modern electronic devices. The second part of my talk will focus on recent achievements in the field of neuromorphic computing realised using exciton-polariton [2-4]. I will provide a brief introduction to the physics of exciton-polaritons and their potential to advance optical computing devices for both classical and quantum computing, particularly within emerging semiconductor material platforms [5-6].
[1] Scientific background: “For foundational discoveries and inventions that enable machine learning with artificial neural networks”, Nobel Lecture (2024)[2] A. Kavokin, T. C. H. Liew, C. Schneider, et al., Nat. Rev. Phys. 4, 435 (2022)
[3] A. Opala, M. Matuszewski, Opt. Mater. Express 13, 2674 (2023)
[4] M. Matuszewski, A. Prystupiuk, A. Opala, Phys. Rev. Applied 21, 014028 (2024)
[5] M. Kędziora, A. Opala, R. Mastria, et al., Nat. Mat. 52, 124 (2024)
[6] A. Opala, K. Tyszka, M. Kędziora, et al., arXiv:2412.10865 (2024)
