How do synapses change a circuit’s response?

For: A learner who understands a membrane signal and wants to connect cells.

A synapse is a site where activity in one cell influences another. That influence is not adequately described by drawing an arrow between two circles. We also need to ask how large the influence is, how long it lasts, where it acts and whether recent activity changes it. The Neuromatch synapses tutorial makes a productive starting distinction between connections with fixed properties and connections whose effects change during use.

First separate transmission from learning. Transmission concerns the immediate effect of a signal passing across a connection. A model can describe that effect without allowing the connection to learn anything. MIT’s synapse lecture places transmission alongside saturation and inhibition. Saturation matters because twice as much input need not produce twice as much response. Inhibition matters because input can change whether, and when, another cell responds.

Location also matters. An input arriving on a branching dendrite must influence the rest of a cell through its electrical properties. The dendrites lecture introduces circuit descriptions and a two-compartment model. This offers a reason to move beyond a neuron represented as a single point: a point model deliberately discards spatial differences that may be central to the question being asked.

Now introduce time at two different scales. Short-term changes can make repeated inputs more or less effective during a sequence. A longer-term plasticity rule instead describes how a connection’s strength changes with experience in the model. The spike-timing-dependent plasticity tutorial explores one such rule, based on the relative timing of activity. It is an example of a learning mechanism, not a universal law for every synapse or a complete explanation of human memory.

Finally, a circuit can change its behaviour even when no individual connection changes strength. Different inputs, combinations of activity and interactions between excitation and inhibition can produce different outputs. The input-correlation tutorial helps separate a connection’s properties from patterns arriving through it. The population-model continuation then asks what interacting groups can do together.

The central habit is to name the proposed change precisely. Does the explanation concern input timing, temporary effectiveness, persistent connection strength or the state of the wider circuit? Calling all of these “rewiring” conceals distinctions that experiments and models need to preserve. This guide develops those distinctions before asking how any of them relate to a particular kind of learning.

Next step

Compare a single-neuron model with a recurrent network and identify which explanation requires the network.

Limits of this route

The Neuromatch materials are teaching models. The inspected sections do not establish one plasticity rule for all cells or a complete biological theory of memory. Source descriptions and tutorial text were read; videos and code were not fully reviewed.