Poster presentation at Cognitive Computing 2018: Unsupervised MNIST Learning in an analog Spiking Neural Network using digital memristive devices

The NBS group led by Elisabetta Chicca is proud to announce that Sören Rüttgers presented the poster "Unsupervised MNIST Learning in an analog Spiking Neural Network using digital memristive devices" at the Cognitive Computing 2018 conference in Hannover, Germany.

Sören presented results from his Bachelor project on a feed-forward spiking neural network incorporating stochastically switching memristive devices as synapses classifying the MNIST dataset of handwritten digits. The input images of the digits were first converted into a spiketrain and fed into the neural network. By not using a teacher signal on the output layer the network was able to learn the ten different digits in an unsupervised way. The stochastic switching behavior of the memristive devices between two stable resistive states was used to emulate the spike-timing-dependent plasticity (STDP) learning rule. A Winner-Take-All (WTA) feature at the output level was used to reinforce the classification output.

With this work the potential of using memristive devices as synapse in spiking neural network simulations is highlighted. Further investigations into different network architectures will pave the way for an energy and area efficient emulation of neural networks.