Redistribution of synaptic efficacy supports stable pattern learning in neural networks

Area of Science:

• Computational neuroscience
• Neural network modeling
• Synaptic plasticity

Background:

• Long-term potentiation (LTP) traditionally assumed to increase synaptic efficacy.
• Markram and Tsodyks observed LTP's efficacy diminishes with higher-frequency pulses, challenging this assumption.
• This frequency-dependent change is termed redistribution of synaptic efficacy (RSE).

Purpose of the Study:

• To propose a computational model explaining Redistribution of Synaptic Efficacy (RSE).
• To investigate RSE as a mechanism for stable, distributed neural network learning.
• To challenge the conventional view of synaptic potentiation as a general gain increase.

Main Methods:

• Developed a computational model with adaptive thresholds instead of multiplicative weights.
• Implemented a distributed instar learning law for monotonic threshold increases.
• Analyzed the model's postsynaptic potential adaptation and frequency-dependent components.

Main Results:

• The model's synaptic balance produced frequency-dependent changes consistent with RSE observations.
• Adaptive thresholds, not multiplicative weights, were central to the model's function.
• RSE was shown to support pattern selectivity and a frequency-independent strengthening component.

Conclusions:

• RSE serves a functional purpose in neural networks for pattern coding.
• The adaptive threshold model supports stable, distributed learning schemes.
• RSE facilitates robust learning capabilities, including fast and stable distributed learning.