Deep neural networks are among the most widely applied machine learning tools showing outstanding performance in a broad range of tasks. In this talk, a method for folding a deep neural network of arbitrary size into a single neuron with multiple time-delayed feedback loops is presented. The network states emerge in time as a temporal unfolding of the neuron's dynamics and by adjusting the feedback-modulation within the loops, the network's connection weights can be adapted. These connection weights are determined, i.e. learned, via a back-propagation algorithm. This approach fully recovers standard Deep Neural Networks (DNN), encompasses sparse DNNs, and extends the DNN concept toward dynamical systems implementations. This new method, which we have called Folded-in-time DNN (Fit-DNN), exhibits promising performance in a set of benchmark tasks.
Meeting ID: 838 2931 8876
Tobias Galla Contact form