Connection of renormalization group with machine learning in spin models

Postgraduate Thesis uoadl:2924555 187 Read counter

Unit:
Κατεύθυνση Πυρηνική Φυσική και Φυσική Στοιχειωδών Σωματιδίων
Library of the School of Science
Deposit date:
2020-10-12
Year:
2020
Author:
Kakampakos Apostolos
Supervisors info:
Φώτιος Διάκονος, Αναπληρωτής Καθηγητής, Τμήμα Φυσικής, ΕΚΠΑ
Original Title:
Σύνδεση ομάδας ανακανονικοποίησης με μηχανική μάθηση σε μοντέλα σπιν
Languages:
Greek
English
Translated title:
Connection of renormalization group with machine learning in spin models
Summary:
Neural networks are the basic tool of analysis in the field of machine learning. Furthermore they have have been used in many problems in physics with great success. But neural networks, especially unsupervised, are considered a black box, namely it is not completely understood how they process received information. Their analysis is very important, since they are used daily in various problems. With this in mind, we applied the Restricted Boltzmann Machines (RBM) to the two dimensional Ising, so as to examine the RBM flow and its fixed point, which seems to be related to the unstable fixed point of the renormalization group. In the beginning we verify that the RBM flow converges to the critical temperature in accordance to bibliography. Furthermore we study other quantities of Ising during the RBM flow. In the end it is attempted to describe the whole procedure using mean field theory.
Main subject category:
Science
Keywords:
neural, RBM, visible, hidden, flow, Ising, critical, mean, theory
Index:
No
Number of index pages:
0
Contains images:
Yes
Number of references:
38
Number of pages:
42
Διπλωματική_Εργασία_Κακαμπάκος_Απόστολος.pdf (1 MB) Open in new window