Related Experiment Videos
Connectionist networks learn to transmit chaos.
G J Mpitsos1, R M Burton, H C Creech
1Mark O. Hatfield Marine Science Center, Oregon State University, Newport, OR 97365.
Brain Research Bulletin
|September 1, 1988
Summary
Connectionist networks can learn and transmit chaotic signals within the nervous system. This research demonstrates how neural networks can process complex chaotic data, advancing our understanding of brain signal transmission.
Area of Science:
- Computational Neuroscience
- Chaos Theory in Biology
Background:
- Some neuronal activity in coordinated motor patterns may involve chaos.
- Biological systems, even simple ones, are challenging to control and model.
Purpose of the Study:
- To investigate if connectionist networks can learn and transmit chaotic signals within the nervous system.
- To explore the capacity of artificial neural networks to process chaotic signals originating from one neural region and being transmitted to another.
Main Methods:
- A simple connectionist network (1 input, 4 hidden, 1 output unit) was trained using chaotic signals.
- Inputs included logistic equations and the Rössler attractor.
- The backpropagation algorithm adjusted synaptic weights based on the error between input and output signals.
Main Results:
- Networks learned to transmit different chaotic attractors when larger error feedback was present.
- Small analog value changes led to high output similarity but poor learning.
- Once trained on one chaotic input, the network could transmit others without synapse modification.
- More hidden units accelerated the learning rate.
Conclusions:
- Connectionist networks demonstrate the potential to learn and transmit chaotic signals, mimicking neural processing.
- The findings suggest a mechanism for how chaotic signals might be processed and relayed within the nervous system.
- Network architecture, particularly the number of hidden units, influences the efficiency of learning chaotic dynamics.