Machine learning (ML) has revolutionized various industries and fundamental sciences by learning important patterns from large datasets. This is similar to how physics predicts the behavior of complex systems through experimental or theoretical data. The relationship between ML and physics is mutually beneficial. ML is used in physics to speed up numerical simulation, design new materials, analyze experimental data and so on. Physics, in turn, is shaping the future of ML through the development of new hardware architectures and theoretical concepts. The thesis focuses on the contributions of physics to ML in hardware, presenting an all-optical neural network with nonlinear activation functions and in theory, proposing a new set of machine learning models called deep self-learning neural...[ Read more ]

Machine learning (ML) has revolutionized various industries and fundamental sciences by learning important patterns from large datasets. This is similar to how physics predicts the behavior of complex systems through experimental or theoretical data. The relationship between ML and physics is mutually beneficial. ML is used in physics to speed up numerical simulation, design new materials, analyze experimental data and so on. Physics, in turn, is shaping the future of ML through the development of new hardware architectures and theoretical concepts. The thesis focuses on the contributions of physics to ML in hardware, presenting an all-optical neural network with nonlinear activation functions and in theory, proposing a new set of machine learning models called deep self-learning neural networks by field theory. The thesis also includes chapter on review of various ML applications in physics.