The expressive power of neural networks in quantum physics
© 2022 EPFL
A collaboration between the CQSL lab at EPFL and the Hebrew University of Jerusalem in Israel has studied the expressive power of neural network representation of quantum states. By means of an efficient mapping of tensor contractions to deep neural networks, the work establishes for the first time the efficient representability of one-dimensional gapped ground states by means of neural quantum states. Other connections to tensor-network states are also discussed.
A recent collaboration between CQSL's head Giuseppe Carleo and computer scientists at the Hebrew University of Jerusalem studies how well artificial neural networks can represent the wave functions of many-body quantum systems. The article, now published in Physical Review B, constructs an efficient mapping between tensor contractions (as used for example in Matrix Product States) and deep feed-forward networks. As a result of this mapping it is established, for example, that neural quantum states can represent efficiently all ground states of gapped one-dimensional hamiltonians.
"Neural tensor contractions and the expressive power of deep neural quantum states", Phys. Rev. B 106, 205136 (2022).