Sessions | Days | Selection | Search | Updates | Downloads | Help
FM: Fall Meeting
FM 79: Entanglement: Neural Networks for Many-Body Dynamics
FM 79.4: Talk
Thursday, September 26, 2019, 14:45–15:00, 2004
Efficient training for neural-network quantum states — •Sheng-Hsuan Lin and Frank Pollmann — Department of Physics, T42, Technische Universität München, James-Franck-Straße 1, D-85748 Garching, Germany
Neural networks have been demonstrated to be a promising approach to represent many-body quantum states. However, this approach suffers from difficulties in optimization for realistic models for two main reasons: Inefficiencies in the Markov chain Monte Carlo sampling procedure and the high cost for the stochastic reconfiguration procedure in variational Monte Carlo method. Recently, it has been shown that neural autoregressive quantum states, which are motivated by the architecture known as autoregressive models, lead to an efficient direct sampling procedure which overcomes the first difficulty. Here we consider the approximated second order method proposed in the machine learning community to investigate the possibility to overcome the second difficulty. We benchmark our algorithm by considering the frustrated J1-J2 model on the square lattice.