Fisher information in scattering problems and neural networks

Abstract

In my talk, I will discuss recent progress in applying the concept of Fisher information to the problem of estimating system parameters in complex scattering environments, such as inside or behind a disordered medium [1,2,3]. We have recently realised that such tools can also be successfully applied to artificial neural networks, in particular to define the performance limit of a network in extracting information from a complex system. [1] Maximum information states for coherent scattering measurements, D. Bouchet, S. Rotter, and A. P. Mosk, Nature Physics 17, 564 (2021). [2] Optimal control of coherent light scattering for binary decision problems, D. Bouchet, L. M. Rachbauer, S. Rotter, A. P. Mosk, and E. Bossy, Phys. Rev. Lett. 127, 253902 (2021). [3] Invariance property of the Fisher information in scattering media, M. Horodynski, D. Bouchet, M. Kühmayer, and S. Rotter, Phys. Rev. Lett. 127, 233201 (2021).