Noise Robust Projection Rule for Hyperbolic Hopfield Neural Networks

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Complex-valued Hopfield neural networks (CHNNs) are multistate models but suffer from low noise tolerance.
• Hyperbolic Hopfield neural networks (HHNNs) offer improved noise robustness.
• Existing HHNN projection rules worsen noise tolerance with more patterns due to self-loops.

Purpose of the Study:

• To enhance the noise tolerance of hyperbolic Hopfield neural networks (HHNNs).
• To address the limitations of the projection rule in HHNNs concerning self-loops and noise sensitivity.

Main Methods:

• Extended the stability condition for self-loops in HHNNs.
• Modified the projection rule for HHNNs to mitigate self-loop issues.
• Evaluated the impact of these modifications on noise tolerance.

Main Results:

• The modified projection rule and extended stability conditions significantly improved HHNN noise tolerance.
• The issue of worsening noise tolerance with increased training patterns was effectively addressed.
• HHNNs demonstrated enhanced performance in the presence of noise.

Conclusions:

• The proposed modifications provide a viable solution for improving HHNN noise tolerance.
• This work advances the development of robust multistate neural network models.
• The enhanced HHNNs are more suitable for applications requiring high noise immunity.