Noise in Neurons and Synapses Enables Reliable Associative Memory Storage in Local Cortical Circuits.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Brain neural networks exhibit remarkable reliability despite inherent noise in signal transmission.
• Sources of noise include presynaptic input errors, synaptic transmission variability, and postsynaptic potential fluctuations.
• The impact of this unreliability on fundamental brain functions like learning and memory remains an open question.

Purpose of the Study:

• To investigate the effects of errors and noise on associative sequence learning in model neural networks.
• To determine if unreliable network activity hinders or aids learning and memory retrieval.
• To explore the relationship between noise, memory capacity, and retrieval efficiency.

Main Methods:

• Analytical and numerical solutions for associative learning problems in networks of inhibitory and excitatory neurons.
• Implementation of a biologically plausible perceptron-type learning rule for loading memory sequences.
• Analysis of network properties under varying levels of noise and error during learning.

Main Results:

• Errors and noise during the learning phase significantly increase the probability of successful memory recall.
• A trade-off exists between the capacity of stored memories and their reliability.
• Noise during learning is essential for achieving optimal retrieval of stored information.

Conclusions:

• Unreliable neural activity, particularly during learning, can paradoxically enhance memory retrieval.
• Networks optimized for associative memory exhibit structural and dynamical features similar to mammalian cortical circuits.
• Predictions are made regarding synaptic plasticity: connections from unreliable neurons may be weakened, while those onto noisy neurons may have lower probability and higher weights.