Figure 6
From: Scalable excitatory synaptic circuit design using floating gate based leaky integrators
Comparison between unsupervised and supervised adaptation cases. (a) Unsupervised competition in the same network as the inset in Fig. 5(a). One of N 1 and N 2 was randomly chosen every time bin (width: 200 ms), and the chosen neuron was given a spiking rate of 20 Hz. Sampled synaptic evolution trajectories are plotted on the phase plane in (b)—each color denotes each sample. The two opposite types of trajectories were observed at random at the equal probabilities. (c) Supervised adaptation by deploying a bias neuron (N sup) that spikes at 80 Hz in sync with N 1. With the aid of N sup, N 1 always wins N 2 as seen on the phase plane in (d). For N sup in sync with N 2, which makes N 2 win N 1, data are shown in (e) and (f).
