TY  - CONF
AU  - Nestler, Sandra
AU  - Keup, Christian
AU  - Dahmen, David
AU  - Gilson, Matthieu
AU  - Rauhut, Holger
AU  - Helias, Moritz
AU  - Bartolomaeus, Wiebke
AU  - Boutaib, Youness
AU  - Bouss, Peter
AU  - Merger, Claudia Lioba
AU  - Fischer, Kirsten
AU  - Rene, Alexandre
AU  - Schirrmeister, Robin
AU  - Ball, Tonio
TI  - Neural networks learning structure in temporal signals
M1  - FZJ-2023-01742
PY  - 2023
AB  - Cortical networks are strongly recurrent, and neurons have intrinsic temporal dynamics. This sets them apart from deep feed-forward networks. Despite the tremendous progress in the application of feed-forward networks and their theoretical understanding, it remains unclear how the interplay of recurrence and non-linearities in recurrent cortical networks contributes to their function. The purpose of this work is to present a solvable recurrent network model that links to feed forward networks. By perturbative methods we transform the time-continuous,recurrent dynamics into an effective feed-forward structure of linear and non-linear temporal kernels. The resulting analytical expressions allow us to build optimal time-series classifiers from random reservoir networks. Firstly, this allows us to optimize not only the readout vectors, but also the input projection, demonstrating a strong potential performance gain. Secondly, the analysis exposes how the second order stimulus statistics is a crucial element that interacts with the non-linearity of the dynamics and boosts performance.
T2  - Seminar Talk, Barak Lab
CY  - , Haifa (Israel)
M2  - Haifa, Israel
LB  - PUB:(DE-HGF)31
UR  - https://juser.fz-juelich.de/record/1006605
ER  -