C3GAN: A brain-inspired memory consolidation for class-incremental learning.

Area of Science:

• Artificial Intelligence
• Neuroscience
• Machine Learning

Background:

• Deep neural networks face catastrophic forgetting, losing prior knowledge when learning new tasks.
• Human memory consolidation involves reactivation, a process explored for artificial systems.
• Existing generative replay methods struggle with complex data and can also suffer from forgetting.

Purpose of the Study:

• To develop a brain-inspired model that addresses catastrophic forgetting in artificial neural networks.
• To emulate human memory consolidation processes for lifelong learning in AI.
• To improve knowledge retention in AI systems without relying on raw data.

Main Methods:

• Proposed C3GAN (Contrastive Clustering and Conditional Generative Adversarial Networks) model.
• Utilized contrastive class structuring for recent memory consolidation (hippocampal mimicry).
• Incorporated a conditional generative adversarial network for long-term knowledge storage (prefrontal cortex mimicry).
• Implemented an amygdala-inspired module for selective replay of challenging classes.

Main Results:

• C3GAN achieved state-of-the-art performance on class-incremental learning benchmarks.
• The model demonstrated effective lifelong memory retention without requiring raw data.
• Brain-inspired mechanisms successfully mitigated catastrophic forgetting.

Conclusions:

• C3GAN offers a novel solution for lifelong memory retention in artificial systems.
• The model successfully emulates key human memory consolidation processes.
• This approach advances the development of AI systems capable of continuous learning.