Complex Learning in Bio-plausible Memristive Networks.

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Computational Neuroscience

Background:

• Existing memristive networks lack internal dynamics (feedforward) or efficient learning algorithms (recurrent), limiting their capabilities.
• Memristor-based neuromorphic systems offer potential for efficient computing but face challenges in complex learning.
• Bio-plausible recurrent networks with internal dynamics are crucial for advanced cognitive functions.

Purpose of the Study:

• To develop a framework supporting complex learning functions in neuromorphic systems.
• To integrate dedicated learning algorithms with bio-plausible recurrent memristive networks.
• To enhance the learning ability of memristor-based neuromorphic computing.

Main Methods:

• Fabrication of iron oxide memristor-based synapses with controllable plasticity and wide dynamic range.
• Construction of a bio-plausible recurrent memristive network incorporating internal dynamics.
• Implementation of the comprehensive recursive least-squares (RLS) learning algorithm for adaptive synaptic weight modification.

Main Results:

• Demonstrated learning of various timing patterns using the proposed framework.
• Successfully learned a complex spatiotemporal pattern of human motor activity.
• Established a functional recurrent memristive network capable of adaptive weight modification.

Conclusions:

• The proposed framework enables complex learning functions in neuromorphic systems.
• Iron oxide memristors provide a robust platform for brain-inspired computing.
• This work advances the exploration of brain-inspired complex learning in neuromorphic engineering.