Snap-drift neural network for self-organisation and sequence learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Traditional neural networks face challenges in self-organisation and sequence learning.
• Modal learning offers a promising approach by combining different learning strategies.

Purpose of the Study:

• To introduce two novel neural networks, the snap-drift self-organising map (SDSOM) and the recurrent snap-drift neural network (RSDNN).
• To evaluate the effectiveness and learning speed of these networks in self-organisation and sequence learning tasks.

Main Methods:

• The snap-drift neural network combines fuzzy AND learning (snap) and Learning Vector Quantisation (drift).
• The SDSOM utilizes a standard SOM architecture with snap or drift weight updates.
• The RSDNN employs a simple recurrent network (SRN) architecture with probabilistic adaptation and reinforcement learning for mode switching.

Main Results:

• Both SDSOM and RSDNN demonstrated effective learning capabilities on benchmark datasets.
• The proposed snap-drift networks exhibited faster learning speeds compared to alternative neural network methods.
• Probabilistic adaptation and reinforcement-based mode switching in RSDNN contributed to improved performance.

Conclusions:

• The novel snap-drift neural networks offer a significant advancement in self-organisation and sequence learning.
• These networks provide a faster and more efficient alternative to existing neural network approaches.
• The findings suggest broad applicability in complex learning tasks requiring self-organisation and temporal processing.