Unformatted text preview: For "batch learning" with all the data included, the learning rule becomes:
M ΔW = k ∑ ( y ( m ) − y ( m ) )x ( m )T where y ( m ) = Wx ( m ) is the output vector.
m =1 When learning stops (ΔW = 0), we have YX T = WXX T (normal equation). Thus W = YX T ( XX T )−1 = YX † if the matrix inverse exists. Computational theories of learning
• Supervised learning:
A teaching signal knows the exact value of the desired
output and corrects the error of the actual output.
Examples: simple and multilayer perceptrons
• Unsupervised learning:
No explicit teaching signal.
Examples: Hebb rule, self-organizing maps
• Reinforcement learning:
A reward signal without knowing the exact output. dependent learning. More precise informa- touch a lever after the appearance of a small established body o
tion about the role played by midbrain do- light. Before training and in the initial 7). From this pers
paminergic activity derives from experiments phases of training, most dopamine neurons mine neurons do
Schultz Behavioral and Brain single 2010, 6:24
in which activity of F...
View Full Document
- Spring '14
- Chemical synapse, Long-term potentiation, NMDA, dopamine neurons