Variable threshold as a model for selective attention, (de)sensitization, and anesthesia in associative neural...

Area of Science:

• Computational Neuroscience
• Artificial Neural Networks
• Cognitive Science

Background:

• Associative neural networks are models of memory and learning.
• Neuronal thresholds are critical for signal processing and network dynamics.
• Noise and pattern activity levels influence network performance.

Purpose of the Study:

• To investigate the impact of variable neuronal thresholds on associative neural network retrieval.
• To analyze how threshold variations affect fixed points and convergence rates under noise.
• To explore the relationship between pattern activity, threshold levels, and network performance.

Main Methods:

• Modified Hebbian learning rule incorporating random pattern activity distributions.
• Analysis of fixed points and convergence rates with variable and constant neuronal thresholds.
• Simulation of network behavior across different noise levels and pattern activity distributions.

Main Results:

• A variable threshold optimizes retrieval for specific pattern activity levels, potentially at the cost of others.
• Constant negative thresholds can enhance overlap for high-activity patterns in high noise, while positive thresholds decrease it.
• Negative thresholds generally yield faster convergence rates than positive thresholds.

Conclusions:

• Neuronal threshold dynamics play a crucial role in selective pattern retrieval and network stability.
• The findings offer insights into cognitive processes like attention and the effects of drugs or injuries on neuronal excitability.
• Dynamic thresholds may underlie phenomena like hysteresis in network recall.