Bicomplex Projection Rule for Complex-Valued Hopfield Neural Networks.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Complex-valued Hopfield neural networks (CHNNs) are multistate models for neural associative memory.
• Existing CHNNs require significant memory due to numerous weight parameters.
• More complex architectures like quaternion- and bicomplex-valued Hopfield networks have been explored but are computationally intensive.

Purpose of the Study:

• To simplify the architecture of complex-valued Hopfield neural networks.
• To reduce the memory resource requirements of CHNNs.
• To maintain or enhance the noise tolerance of CHNNs.

Main Methods:

• Introduction of a bicomplex projection rule derived from the decomposition of bicomplex-valued Hopfield neural networks.
• Application of this rule to reduce the number of weight parameters in CHNNs.
• Validation through computer simulations comparing noise tolerance with other advanced Hopfield network models.

Main Results:

• The bicomplex projection rule significantly reduces the number of weight parameters in CHNNs.
• Computer simulations demonstrate that CHNNs using the bicomplex projection rule exhibit noise tolerance equal to or better than quaternion- and bicomplex-valued Hopfield networks.
• The projection rule for hyperbolic-valued Hopfield neural networks in synchronous mode also maintains high noise tolerance.

Conclusions:

• The proposed bicomplex projection rule offers an effective method for simplifying CHNN architectures and reducing memory demands.
• This simplification does not compromise, and can even improve, the network's noise tolerance.
• The findings suggest a promising direction for developing more efficient and robust neural associative memory models.