Implementing artificial neural networks through bionic construction.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Biomimicry

Background:

• Biological neural networks utilize heredity or learning.
• Traditional artificial neural networks (ANNs), particularly Deep Learning Neural Networks (DNNs), rely on extensive training and learning.
• Existing DNNs often exhibit redundancy, increasing training effort and computational resource demands.

Purpose of the Study:

• To propose a bionic method for implementing ANNs through construction rather than training.
• To design ANN hierarchy based on functional analysis and modular design.
• To validate the proposed method using the Drosophila visual neural network.

Main Methods:

• A bionic approach for ANN implementation via construction.
• Hierarchical network design based on functional requirements.
• Modular design for each hierarchy.
• Testing with the Drosophila visual neural network.

Main Results:

• The developed bionic artificial neural network functions as a bionic compound eye.
• The network successfully achieves object and movement detection.
• Performance shows improvements in certain properties compared to biological Drosophila compound eyes.

Conclusions:

• The construction-based bionic approach offers an alternative to traditional DNN training.
• This method can lead to more efficient and less redundant ANNs.
• The Drosophila visual system serves as a viable model for bionic ANN development.