Constructing multilayered neural networks with sparse, data-driven connectivity using biologically-inspired,...

Area of Science:

• Computational Neuroscience
• Neuroscience
• Artificial Intelligence

Background:

• Brain complexity necessitates intrinsic self-regulating mechanisms for construction and control.
• Synaptogenesis, the formation of new neural connections, is crucial for neuronal connectivity and brain performance.
• Adaptive synaptogenesis networks integrate synaptogenesis, synaptic modification, and shedding for sparse network construction.

Purpose of the Study:

• To incorporate novel neuroscientific observations into adaptive synaptogenesis models.
• To investigate the trade-off between brain performance and energy expenditure.
• To develop a computational model that simulates brain development and function.

Main Methods:

• Incorporation of multiple layers, neuron survival/death based on information transmission, and bigrade growth factor signaling.
• Development of adaptive synaptogenesis networks with enhanced features.
• Simulations to analyze network performance (information loss, classification errors) and energy expenditure (neuron count).

Main Results:

• Identified the critical role of intermediate neural layers in regulating synaptogenesis and neuron elimination.
• Demonstrated performance and energy-saving benefits from delayed synaptogenesis in subsequent layers.
• Showcased how neuron elimination in preceding layers enhances energy savings and code compression without significant performance degradation.

Conclusions:

• The refined adaptive synaptogenesis model effectively balances performance and energy efficiency.
• Strategic control over synaptogenesis timing and neuron elimination are key to optimizing neural network function.
• This research provides insights into efficient brain-like network design for computational applications.