Neurons with hysteresis from a network that can learn without any changes in synaptic connection strengths
A neural network concept derived from an analogy between the immune system and the central nervous system is outlined. The theory is based on a neuron that is slightly more complicated than the conventional McCullogh-Pitts type of neuron, in that is exhibits hysteresis at the single cell level. This added complication is compensated by the fact that a network of such neurons is able to learn without the necessity for any changes in synaptic connection strengths. The learning occurs as a neural consequence of interactions between the network and its environment, with environmental stimuli moving the system around in an N-dimensional phase space, until a point in phase space is reached such that the system's responses are appropriate for dealing with the stimuli. Due to the hysteresis associated with each neuron, the system tends to stay in the region of phase space where it is located. The theory includes a role for sleep in learning. 18 refs., 2 figs.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States); British Columbia Univ., Vancouver (Canada); Lakehead Univ., Thunder Bay, Ontario (Canada). Dept. of Mathematics
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 5618182
- Report Number(s):
- LA-UR-86-1449; CONF-8604173-2; ON: DE86010205
- Resource Relation:
- Conference: Neural nets and computation conference, Snowbird, UT, USA, 1 Apr 1986; Other Information: Portions of this document are illegible in microfiche products
- Country of Publication:
- United States
- Language:
- English
Similar Records
Synaptic connections between the hindwing stretch receptor and flight motor neurones in the locust revealed by double cobalt labelling for electron microscopy
Superconducting neural networks with disordered Josephson junction array synaptic networks and leaky integrate-and-fire loop neurons