Questions: Hebbian Learning Mechanisms

Question 1

A simulated neural network applies only Hebbian potentiation: active synapses strengthen whenever pre- and postsynaptic neurons fire together. After many learning trials, all synaptic weights approach their maximum value and the network loses its ability to discriminate between inputs. What is missing from this model?

• A. Spike-timing-dependent plasticity, which introduces a delay between pre- and postsynaptic firing
• B. Homeostatic mechanisms such as synaptic scaling or heterosynaptic depression to prevent runaway potentiation
• C. AMPA receptor trafficking, which is required for synaptic strengthening to occur physically
• D. Inhibitory interneurons, whose absence allows unconstrained excitatory potentiation

Answer: (B)

Pure Hebbian learning is inherently unstable: strengthening a synapse increases the postsynaptic neuron's firing, which increases its correlation with its inputs, which further strengthens the synapse — a runaway positive feedback loop. Real neural circuits prevent this through homeostatic mechanisms. Synaptic scaling globally adjusts all of a neuron's synapses to maintain stable mean firing rates. Heterosynaptic depression weakens inactive synapses when nearby active ones are strengthened, introducing competition. Without these, the Hebbian rule drives all weights to saturation. Option A (STDP) describes a form of Hebbian plasticity rather than a corrective mechanism. AMPA receptor trafficking (C) is the molecular implementation of Hebbian strengthening, not a stabilizing mechanism.

Question 2

NMDA receptors function as 'coincidence detectors' in Hebbian learning. What specific requirement makes them suited for this role?

• A. NMDA receptors are expressed only at synapses that are newly formed, making them sensitive to developmental activity patterns
• B. NMDA receptors require both glutamate binding (indicating presynaptic activity) AND postsynaptic depolarization (to remove the Mg²⁺ block) before they open and admit Ca²⁺
• C. NMDA receptors are activated exclusively by high-frequency stimulation, which is the signature of correlated neural firing
• D. NMDA receptors detect the time since last presynaptic firing and open only after a critical interval, enabling temporal specificity

Answer: (B)

The magnesium block is the molecular mechanism of coincidence detection. At resting membrane potential, Mg²⁺ ions physically block the NMDA channel pore even when glutamate is bound. Only when the postsynaptic membrane is sufficiently depolarized (by AMPA receptor activation or back-propagating action potentials) is the Mg²⁺ expelled, allowing the channel to pass Ca²⁺. This means the NMDA receptor opens only when two conditions are met simultaneously: presynaptic glutamate release AND postsynaptic depolarization. It is a molecular AND gate that detects coincident activity — exactly Hebb's rule implemented in biochemistry. The resulting Ca²⁺ influx triggers AMPA receptor insertion, strengthening the synapse.

Question 3

Hebbian plasticity can produce synaptic depression (LTD) as well as potentiation (LTP), depending on the temporal relationship between pre- and postsynaptic firing.

• T. True
• F. False

Answer: True

Spike-timing-dependent plasticity (STDP), a specific form of Hebbian plasticity, demonstrates that the sign of synaptic change depends on the order and timing of pre- and postsynaptic spikes. When a presynaptic spike precedes a postsynaptic spike within ~20ms, the synapse potentiates (LTP — consistent with 'A causes B'). When the postsynaptic spike precedes the presynaptic spike, the synapse depresses (LTD — the timing is inconsistent with A causing B). The depression window in STDP is part of what prevents runaway potentiation and provides a form of competition between synapses. The misconception that Hebbian learning only produces potentiation overlooks that depression also requires correlated activity — just in the opposite temporal order.

Question 4

Hebbian learning is self-stabilizing: as synaptic strengthening increases postsynaptic firing, the neuron's threshold automatically rises to prevent further potentiation.

• T. True
• F. False

Answer: False

This is the opposite of what pure Hebbian learning predicts. The Hebbian rule contains no intrinsic negative feedback. Strengthening a synapse increases postsynaptic firing, which increases the correlation between pre- and postsynaptic activity, which triggers further strengthening — a positive feedback loop leading to runaway potentiation. Stability requires *separate* homeostatic mechanisms external to the Hebbian rule itself: synaptic scaling (which adjusts all synapses globally to preserve mean firing rate), heterosynaptic depression (which introduces lateral competition), and intrinsic excitability regulation. The Hebbian rule is 'fire together wire together' — there is no mechanism within that rule to stop the wiring from becoming absolute.

Question 5

Explain why pure Hebbian learning is unstable, and describe at least one mechanism the brain uses to prevent runaway potentiation.

Model answer: Pure Hebbian learning is unstable because it contains only positive feedback: strengthening a synapse causes the postsynaptic neuron to fire more often, which increases its correlation with the presynaptic neuron, which further strengthens the synapse. Without any counterbalancing mechanism, all active synapses drive toward maximum strength, saturating the neuron's response and eliminating its ability to discriminate between different input patterns. The brain uses several complementary mechanisms to stabilize Hebbian plasticity. Synaptic scaling globally scales all of a neuron's synapses up or down to maintain a target mean firing rate — if a neuron fires too much, all its synapses weaken proportionally. Heterosynaptic depression weakens inactive synapses when neighboring active synapses are potentiated, introducing competition. The depression window in spike-timing-dependent plasticity provides temporal specificity, weakening synapses when postsynaptic firing precedes presynaptic input. These mechanisms operate in addition to the Hebbian rule, not within it, providing the complementary constraints that allow correlation-based learning to be selective rather than totalizing.

The key insight is that Hebbian learning describes a powerful but incomplete rule. Its instability is not a bug to be patched but a fundamental property that explains why the brain requires complementary homeostatic mechanisms. Understanding the failure mode of pure Hebbian learning clarifies why synaptic scaling, heterosynaptic depression, and STDP's depression component are necessary features of real neural circuits, not optional add-ons.