Figure 3
From: Using Memristors for Robust Local Learning of Hardware Restricted Boltzmann Machines
Examples of hidden features extracted by, from left to right: standard RBM (trained with a learning rate of 0.05), a memristive Discriminative RBM (trained under Cst with β = 0.005, Δt/Δtmax = 1/1000), another memristive Discriminative RBM (trained under Cst with β = 3, Δt/Δtmax = 1/5000). Each gray-scale picture represents the values of the 784 weights connecting the visible layer to a given hidden unit.
