Dendrites endow artificial neural networks with accurate, robust and parameter-efficient learning.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Deep learning (DL) algorithms, based on artificial neural networks (ANNs), excel at complex tasks but are parameter-heavy, energy-intensive, and prone to overfitting.
• Biological brains solve similar problems with remarkable efficiency, suggesting potential for bio-inspired AI architectures.

Purpose of the Study:

• To introduce a novel ANN architecture inspired by the structured connectivity and restricted sampling of biological dendrites.
• To evaluate the performance and efficiency of this new dendritic ANN architecture compared to traditional ANNs.

Main Methods:

• Developed a new ANN architecture incorporating dendritic properties like structured connectivity and restricted sampling.
• Tested the dendritic ANNs on various image classification tasks.
• Compared the performance, parameter efficiency, and robustness to overfitting against traditional ANNs.

Main Results:

• Dendritic ANNs demonstrated increased robustness against overfitting.
• These novel ANNs outperformed traditional ANNs on image classification benchmarks.
• The dendritic architecture required significantly fewer trainable parameters than conventional ANNs.

Conclusions:

• Incorporating biological dendritic properties enhances ANN precision, resilience, and parameter efficiency.
• The learning strategy in dendritic ANNs, where nodes respond to multiple classes, differs from traditional class-specific approaches.
• This research highlights the potential of bio-inspired designs to improve deep learning algorithms.