Running applications with millions of neurons on Darwin3 (IMAGE)
Caption
Figure (a) illustrates a neural network consisting of five spiking VGG-16 as the core components. The network is highly adaptable to various artificial intelligence scenarios. Figure (b) and (c) illustrate a neural network directly on-line trained to solve mazes by mapping the maze onto a set of neurons. Excitatory neurons represent free-walking points, while inhibitory neurons represent obstacles. During learning, synaptic weights increase based on specific rules, and inhibitory neurons maintain inhibition. Once the model has learned, it can quickly find the path by observing the direction along which spikes propagate.
Credit
©Science China Press
Usage Restrictions
Use with credit.
License
Original content