Menu

ExtrapolationPerformance

Matus Uzak

Extrapolation Performance

The following images display responses of an output neuron of an MLP network to data from visualization set. A visualization set includes both the training set and the testing set. It also includes additional data extending the original feature space in both axes. Extrapolation performance aids the teacher in understanding of the knowledge formation process. This positions the human impression into control of the training process.

The are two options in means of interaction. A teacher can either stop the training process, reinitialize weights and start training from scratch. Or fine tune the network response to acquire better approximation by resetting weights to selected neurons.

Spirals Task

spiral1
spiral2
spiral_clear2
spiral_x1
spiral_x2
spiral_x3
spiral_x4
spiral_x5
spiral_x6
spiral_x7
spiral_x9


Related

Wiki: Home