Adaptation of the relaxation method for learning in bidirectional associative memory

Area of Science:

• Artificial Intelligence
• Machine Learning
• Neural Networks

Background:

• Discrete Bidirectional Associative Memory (BAM) networks are crucial for pattern recognition and associative recall.
• Existing iterative learning algorithms often rely on gradient descent, which can be computationally intensive and sensitive to parameters.
• There is a need for efficient and robust learning algorithms for BAMs that ensure complete pattern recall.

Purpose of the Study:

• To introduce and describe a novel iterative learning algorithm named PRLAB.
• To demonstrate PRLAB's capability to guarantee the recall of all training pairs in discrete BAMs.
• To analyze the performance, speed, parameter insensitivity, and scalability of PRLAB.

Main Methods:

• PRLAB is an iterative learning algorithm adapted from the relaxation method for solving systems of linear inequalities.
• It is applied to discrete bidirectional associative memory (BAM) networks.
• The algorithm's performance is evaluated through extensive analysis, including learning speed and scalability tests.

Main Results:

• PRLAB guarantees 100% recall of all training pairs for discrete BAMs.
• The algorithm is significantly faster than existing methods, averaging only 20 epochs for 200 patterns in a 200-200 BAM.
• PRLAB demonstrates high insensitivity to initial configurations and learning parameters.
• The algorithm exhibits excellent scalability, maintaining high performance with increasing network size and pattern load.

Conclusions:

• PRLAB offers a highly effective and efficient solution for training discrete BAM networks.
• Its non-gradient-descent approach and robustness make it a valuable advancement in associative memory research.
• The algorithm's speed, scalability, and guaranteed recall performance position it for practical applications in large-scale systems.