Training a neural network to find the way to a certain location
Green markers are locations an agent can start from. The Red marker is the destination point. The blue arrows mark the directions an agent would proceed when at that position.
The training of the network is done using backpropagation with momentum.
To specify the network structure enter the numbers of neurons in the hiddenlayers in comma separated form. like "20,30" for two hidden layers with 20 and 30 neurons. For the changes to take effect you need to press the init button.
When the network is well trained one should be able to follow the arrows from each start location to the destination point
When the init button is pressed, the network matrices will be initialized with random values between -1 and 1
The displayed error is the average of the squares of the errors used for the backpropagation during the last epoch.
Dijkstra's algorithm is used to create the training data for the map.
Output examples:
Back to main page