Bio-Inspired Architectures Substantially Reduce the Memory Requirements of Neural Network Models

Area of Science:

• Neuroscience
• Computational Biology
• Artificial Intelligence

Background:

• The cricket cercal sensory system detects environmental stimuli like air currents for escape behavior.
• Understanding this system can inform the development of efficient neural network models.

Purpose of the Study:

• To propose a neural network model inspired by the cricket's cercal sensory system for escape response.
• To compare the efficiency and interpretability of this bio-inspired model against generic deep learning models.

Main Methods:

• Developed a neural network model based on physiological and anatomical studies of the cricket sensory system.
• Compared the parameter efficiency of the bio-inspired model with a universal approximation (generic deep learning) model.

Main Results:

• The bio-inspired model achieved comparable performance to generic models but required 1-2 orders of magnitude fewer parameters.
• The proposed model's architecture, based on logical neuron relations, enhances interpretability.

Conclusions:

• Bio-inspired architectural motifs from animal nervous systems can lead to memory-efficient neural network models.
• This approach offers a promising direction for developing interpretable and efficient AI systems.