Spatiotemporal discrimination in attractor networks with short-term synaptic plasticity

Area of Science:

• Computational neuroscience
• Cognitive modeling
• Neural networks

Background:

• Neural networks are fundamental to brain function.
• Understanding how the brain processes sequential information is a key challenge.
• Dynamic synapses and network states are crucial for neural computation.

Purpose of the Study:

• To investigate how randomly connected attractor networks with dynamic synapses process sequential stimuli.
• To determine if such networks can model cognitive phenomena like primacy and recency.
• To explore the potential of these networks as general computational engines for cognitive tasks.

Main Methods:

• Simulated a randomly connected attractor network with bi-stable units and dynamic synapses.
• Introduced sequences of stimuli to the network and analyzed its state transitions.
• Assessed the network's ability to encode sequence identity, recall, and make evidence-based decisions.

Main Results:

• The network successfully discriminated between similar stimulus sequences.
• Network dynamics exhibited primacy and recency effects consistent with human recall data.
• The network could retain information for binary choices based on stimulus presentation frequency.

Conclusions:

• Dynamically connected attractor networks offer a plausible neural basis for sequence discrimination and memory.
• These networks can model key aspects of human cognitive performance, including memory biases and decision-making.
• Such networks represent versatile computational tools for understanding diverse cognitive functions.