PSO-based cloning template design for CNN associative memories

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Cellular Neural Networks (CNNs) are powerful computational models.
• Designing robust CNNs with specific functionalities, like associative memories, remains challenging.
• Existing synthesis methods may lack robustness to parameter variations.

Purpose of the Study:

• To propose a novel synthesis procedure for cellular neural networks (CNNs).
• To develop CNNs capable of space-invariant cloning using evolutionary algorithms.
• To ensure the synthesized networks are robust to parametric variations and noisy inputs.

Main Methods:

• Utilized particle swarm optimization (PSO), an evolutionary algorithm, for CNN synthesis.
• Applied PSO to design CNNs for associative memory applications.
• Incorporated robustness to parametric variations into the design algorithm.

Main Results:

• Successfully synthesized cellular neural networks (CNNs) with space-invariant cloning templates.
• Demonstrated the effectiveness of the particle swarm optimization (PSO) based design.
• Numerical results confirmed good performance in recalling patterns from noisy data.

Conclusions:

• The proposed synthesis procedure offers an effective method for designing robust CNNs.
• Particle swarm optimization (PSO) is a viable approach for CNN synthesis, particularly for associative memories.
• The developed CNNs exhibit resilience to noise, improving their practical applicability.