Also Influence Based Learning — is a new learning method, or learning algorithm, that has been discovered as part of the Netlab development effort. It is based on the idea that pre-synaptic neurons can trust that forward (post-synaptic) neurons are learning (they are, after all, neurons).

By assuming that post-synaptic neurons are learning, pre-synaptic neurons are able to determine which forward neurons to make or strengthen connections with, based on how much influence they are exercising over current responses. This is, in fact, how and why influence learning works.

Because influence over current responses is determined during signal propagation, there is no need to calculate an error value at each output neuron and propagate it backwards through the network. This, in turn, eliminates the need to restrict networks to layered —feed-forward-only— topologies.

Influence learning is a Netlab™ term with no direct analog in the study of biological neural networks. It is, however, loosely based on known biological observations.





. . . . . . .
Neuron Body Values (Some)

NBValue	Description
Output Influenced (OI)	Output Influenced
OI_E	Output Influenced from Excitatory Inputs
OI_I	Output Influenced from Inhibitory Inputs
Input Influenced (II)	Input Influenced
II_E	Input Influenced from Excitatory Inputs
II_I	Input Influenced from Inhibitory Inputs

. . . . . . .
Resources

• How our brains are biologically tuned to be influenced by confident people
  Interesting higher-level evidence for the theory underlying this algorithm.

• From this site:
  • Introducing Influence Learning
  • Biological Underpinnings of Influence Learning
  • Neural Networks Backgrounder: Ce n'est pas une l'histoire
  • Influence Learning Gets A Patent