Geon The Mechanism of Associative Learning Topics


Introduction to Learning

Learning, as first recognized by Ivan Pavlov, is basically a process that associates two events. In our experience, rain always comes with dark cloud. The two events are associated in our brain. Then the dark cloud alone will very likely trigger the memory of rain before it actually happens. The association between dark cloud and rain allows us to avoid unpleasant events well ahead. This capability is essential for survival. When one learns to play chess, he should first memorize the patterns which will immediately lead to the capture of the king. Let us call these patterns {A}. After a few weeks of experience (learning), he may find that some patterns, called {B}, could lead to {A}. The association between {A} and {B} also enables the player to better plan his next move before too late. After years of experiences, he may discover that some patterns {D} are likely to result in patterns {C}, and {C} may lead to {B}. Therefore, a novice player can see only a couple of steps ahead, but an experienced player can see many steps in advance.

A major goal of physical sciences is to find the equations associating the state of a system at a given time t and its next moment, t + dt. The solutions of these equations (known as differential equations) give the entire time course about the system such that physicists can predict the future, if the present information (initial conditions) is known. For instance, meteorologists have been able to develop a set of equations describing the time courses of atmospheric variables (temperature, pressure, etc.). From the present information provided by satellites, they can predict rain several days before clouds turn dark.


Figure 1. The human brain. Hippocampus is a major area for learning and memory. [Source: NIH]

In our brain (Figure 1), the hippocampus is a major area for learning and subsequent formation of memory. In 1950s, a few patients were treated for epilepsy by removing part of the hippocampus. After the surgery, the patients still remembered their names, the events of their childhood, and the language learned before the surgery. However, they could not form new memory. Although they could talk normally with their relatives and friends, but they forgot such events in the next day. The early stage of Alzheimer's disease is characterized by the same symptom, because the hippocampus is damaged early in the progression of Alzheimer's disease.

Cellular Level Description

Recalling that learning is a process that associates two events. In the hippocampus, events are represented by a population of neurons, each may be either "excited" or "resting". A particular event is represented by a particular set of neurons in the "excited" state. Suppose the number of neurons involved in the representation is n, then mathematically an event can be denoted by a vector with dimension n,

X = [x1, x2, x3, .....xn]

where xi (i = 1 - n) is either "0" (resting) or "1"(excited). A very simplified example is depicted in Figure 2A where events are represented by only five neurons (n = 5). Excitation of neurons #2 and #4 is assumed to produce the perception of "dark cloud" while firing of neurons #1 and #5 represents "rain".


Figure 2. A very simplified illustration for associative learning. Only five neurons are assumed to represent events and each neuron contains only one spine.
(A) Dark cloud is represented by the excitation of neurons #2 and #4; rain is represented by firing of neurons #1 and #5.
(B) The synaptic connection between presynaptic neurons (upper row) and postsynaptic neurons (lower row). The yellow line indicates strong synaptic connection after learning, but in the absence of any stimulus, all neurons are in the resting state.
(C) When the dark cloud appears in the sky, the sensory input causes neurons #2 and #4 to fire. This in turn leads to the excitation of neurons #1 and #5 because of strong synaptic connection. The firing of #1 and #5 recalls rain.

The association between two events is accomplished by the synaptic connection among these neurons. In most cases, a synapse is formed between the axon terminal of a neuron (presynaptic) and the dendrites of another neuron (postsynaptic). The contact point on dendrites is typically a raised structure called "dendritic spine". The neural network of the hippocampus is extremely complex. Each neuron may connect up to thousands of other neurons. For simplicity, we assume that a neuron contains only one dendritic spine. Before learning, there was no synaptic connection among these neurons so that the excitation of a neuron does not affect other neurons. During the learning process, synapses are being modified, which may either potentiate or depress the synaptic strength. If the synaptic connection between two neurons is sufficiently strong, excitation of the presynaptic neuron may trigger excitation of the postsynaptic neuron.

In Figure 2B, the yellow line indicates strong synaptic connection between a presynaptic neuron (in upper row) and a postsynaptic neuron (in lower row) after learning. Hence, the perception of "dark cloud" resulting from excitation of neurons #2 and #4 is likely to cause neurons #1 and #5 to fire and recall "rain" (Figure 2C). We see that the association between two events boils down to the link between presynaptic and postsynaptic neurons via synaptic modification. Since events are associated when they appear at the same time, synaptic change should occur only when both presynaptic and postsynaptic neurons are excited simultaneously.

The Role of NMDA Receptors

How can the nervous system implement associative learning? The answer lies in the NMDA receptor (NMDAR), which is a type of ligand-gated ion channels. Unlike other ligand-gated ion channels, whose activation (opening) requires only the binding of neurotransmitters, the activation of NMDAR requires two events: binding of glutamate (a neurotransmitter) and relief of Mg2+ block. NMDARs are located at the postsynaptic membrane. At the resting membrane potential, NMDARs are blocked by Mg2+ ions. If the membrane potential is depolarized due to excitation of the postsynaptic neuron, the outward depolarizing field may repel Mg2+ out of the pore. On the other hand, binding of glutamate can open the intrinsic gate of NMDARs (via unknown mechanisms). In the normal physiological process, glutamate is released from the axon terminal when the presynaptic neuron is excited. Relief of Mg2+ block is due to excitation of the postsynaptic neuron. Therefore, simultaneous excitation of both presynaptic and postsynaptic neurons may open the NMDA receptors.

When the NMDA receptor is open, it is permeable to Ca2+ ions which control the activities of various enzymes. The entry of Ca2+ ions into the postsynaptic neuron can then modify the strength of synaptic connections, resulting in long-term potentiation or depression. Subsequent processes may further produce long lasting memory traces (see Born to Forget, Die to Remember).


Author: Frank Lee
First published:
  September, 2009 (on
  September, 2017 (on