Analytical derivatives of neural networks

Simon Verret, Reza Nourafkan, Quinton Weyrich +2 more

Wojciech Rządkowski

In this Thesis, I study composite quantum impurities with variational techniques, both inspired by machine learning as well as fully analytic. I supplement this with exploration of other applications of machine learning, in particular artificial neural networks, in many-body physics. In Chapters 3 and 4, I study quasiparticle systems with variational approach. I derive a Hamiltonian describing the angulon quasiparticle in the presence of a magnetic field. I apply analytic variational treatment to this Hamiltonian. Then, I introduce a variational approach for non-additive systems, based on artificial neural networks. I exemplify this approach on the example of the polaron quasiparticle (Fröhlich Hamiltonian). In Chapter 5, I continue using artificial neural networks, albeit in a different setting. I apply artificial neural networks to detect phases from snapshots of two types physical systems. Namely, I study Monte Carlo snapshots of multilayer classical spin models as well as molecular dynamics maps of colloidal systems. The main type of networks that I use here are convolutional neural networks, known for their applicability to image data.

Stefanie Czischek

Tariq Mahmood Khan, Antonio Robles‐Kelly

Shuang Cong, Jinmin Yang, Sajede Harraz

A 35 input/single output quantum phase neural network (QPNN) is proposed to recognize the English letters' A 'and' B 'composed 7*5 of pixels. Each node of the proposed QPNN has the input with 0 or 1. Through normalization and quantum phase encoding, each set of 35 with 0 or 1 digital inputs is encoded into a quantum superposition state described by 35 quantum phases. Then, a quantum rotation gate is used to introduce adjustable phase, and a controlled NOT gate is used to entangle the relative phase between the two inputs. The relationship between the input and output of the network is derived. This paper also derived the phase weight learning training algorithm with adaptive learning rate. The solutions with the recognition probability of 1 is designed, and the analytical expressions for all adjustable phase solutions with zero errors are not unique. The experimental performance is verified on the Qiskit platform. This paper uses phase drive to convert the probability calculation into an analytical polynomial function with adjustable rotation angles in the network, avoiding the bottleneck of generating all 2^35 complex exponential amplitudes and providing an effective way to solve the "exponential wall" problem, which provides a new implementation solution for the practical application of multi-input recognition problems.

Yoshihiro Michishita

Machine learning with neural networks is now becoming a more and more powerful tool for various tasks, such as natural language processing, image recognition, winning the game, and even for the issues of physics. Although there are many studies on the application of machine learning to numerical calculation and the assistance of experimental detection, the methods of applying machine learning to find the analytical method are poorly studied. In this letter, we propose methods to use machine learning to find the analytical methods. We demonstrate that the recurrent neural networks can ``derive'' the Floquet-Magnus expansion just by inputting the time-periodic Hamiltonian to the neural networks, and derive the appropriate rotating frame in the periodically-driven system. We also argue that this method is also applicable to finding other theoretical frameworks in other systems.