Autoshaping and automaintenance: a neural-network approach

Area of Science:

• Neuroscience
• Computational Psychology
• Animal Behavior

Background:

• Autoshaping and automaintenance are key learning phenomena.
• Existing models often distinguish between operant and respondent learning.
• Neural systems, including hippocampal and dopaminergic pathways, are implicated in learning.

Purpose of the Study:

• To present an interpretation of autoshaping and automaintenance using a unified neural-network model.
• To investigate the role of response feedback in these learning processes.
• To explore implications for the operant-respondent distinction and behavior-neuroscience relationships.

Main Methods:

• Developed a neural-network model that does not distinguish between operant and respondent learning.
• Simulated autoshaping and automaintenance using an A-B-A design with feedforward architectures.
• Varied network connectivity and the intensity of simulated sensory feedback from responding.

Main Results:

• A fully connected network with maximal response feedback produced substantial autoshaping and automaintenance.
• Eliminating response feedback significantly interfered with both autoshaping and automaintenance.
• Intermediate levels of response feedback resulted in intermediate levels of autoshaping and automaintenance.

Conclusions:

• Response feedback is a critical factor in autoshaping and automaintenance.
• A unified neural network model can account for these learning phenomena without distinguishing between learning types.
• The findings support a closer integration of behavioral and neuroscientific approaches to learning.