Incorporating rapid neocortical learning of new schema-consistent information into complementary learning systems...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The complementary learning systems theory posits that the hippocampus and neocortex avoid rapid integration of arbitrary new information to prevent catastrophic interference with existing knowledge.
• Prior studies suggested neocortical circuits can learn new associations consistent with existing knowledge, challenging the initial theory.
• Artificial neural network models have been used to explore these learning mechanisms.

Purpose of the Study:

• To re-evaluate the complementary learning systems theory using extended neural network simulations.
• To investigate the role of prior knowledge consistency in learning and interference.
• To explore how the rate of change in neural connections depends on existing knowledge.

Main Methods:

• Extended simulations based on the McClelland et al. (1995) neural network model.
• Simulations incorporating new information consistent and inconsistent with prior knowledge.
• Development of a more biologically plausible network architecture to model cortical learning.

Main Results:

• Simulations demonstrated rapid learning of consistent new information without catastrophic interference.
• Catastrophic interference occurred primarily when learning inconsistent new information rapidly.
• The rate of change of connections in both artificial and real neural networks is strongly dependent on prior knowledge.

Conclusions:

• The findings support the complementary learning systems theory, highlighting the crucial role of knowledge consistency.
• Neocortical learning mechanisms, as modeled by neural networks, can rapidly acquire consistent new associations.
• The rate of synaptic plasticity is modulated by existing knowledge, influencing learning and memory consolidation.