Design of general bidirectional associative memories with improved recall capability

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Theory

Background:

• Bidirectional Associative Memories (BAMs) are neural networks used for associative recall.
• Enhancing the recall capability of BAMs is crucial for their practical applications.
• The concept of Hamming balls is relevant for analyzing the separability and performance of associative memory models.

Purpose of the Study:

• To introduce the concept of qth order Hamming balls for General Bidirectional Associative Memories (GBAMs).
• To propose a novel technique for addressing the separability problem in qth order Hamming balls.
• To demonstrate the improved recall capability of GBAMs using the proposed method.

Main Methods:

• The study defines and utilizes the concept of qth order Hamming balls.
• A Minimum Overlap Algorithm (MOA) is proposed to manage the separability of these Hamming balls.
• Simulations were conducted to evaluate the performance of the proposed technique.

Main Results:

• The proposed method effectively deals with the separability problem of qth order Hamming balls.
• Simulation results indicate a significant enhancement in recall capability.
• The recall capability achieved surpasses that of conventional methods.

Conclusions:

• The qth order Hamming ball concept offers a way to enhance GBAM recall.
• The Minimum Overlap Algorithm (MOA) provides an effective solution for separability issues.
• The proposed approach represents a notable advancement in associative memory performance.