Growing dendrites enhance a neuron's computational power and memory capacity.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Neocortical pyramidal neurons exhibit dendritic plasticity, with significant growth during early human development.
• Individual dendrites can generate neuronal spikes independently, suggesting localized processing capabilities.

Purpose of the Study:

• To investigate the neurocomputational advantages and limitations of a novel algorithm combining dendritogenesis and supervised adaptive synaptogenesis.
• To explore the potential of this algorithm for enhancing memory capacity and generalization in artificial neurons.

Main Methods:

• Development of a local, stochastic algorithm inspired by Hebbian developmental theory.
• Integration of dendritogenesis (dendrite growth) with supervised adaptive synaptogenesis (synapse formation).
• Evaluation of the algorithm's performance on classification tasks with input perturbations.

Main Results:

• Neurons developed using the algorithm demonstrated enhanced memory capacity and resistance to catastrophic forgetting.
• The algorithm enabled individual dendrites to form unsupervised feature-clusters, unmixing mixture distributions.
• Error-free classification was achieved even with up to 40% input perturbations.

Conclusions:

• The novel stochastic algorithm facilitates unsupervised dendritic development into feature-clusters, aiding in unmixing mixture distributions.
• This generative model offers significant advantages for generalization and extrapolation beyond learned data, ideal for complex decision-making.