Three-Factor Learning¶
A synaptic learning-rule structure in which presynaptic activity, postsynaptic state, and an additional modulatory signal jointly govern plasticity.
Core Idea¶
A three-factor learning rule changes a synapse according to its presynaptic activity, its postsynaptic state, and a third modulatory signal. The first two are local to the connection; the third can carry feedback such as reward, error or novelty. In delayed variants, a temporary eligibility trace remembers which synapses were recently active until the modulator arrives. The third factor need not always be dopamine, global, or represented by exactly one multiplication formula.[^ref-325874ea5d3c]
In delayed, trace-based versions, this combination can help a later outcome affect recently relevant connections without giving every synapse a separate error signal. The three-input rule need not use that particular sequence, and it does not guarantee perfect credit assignment or learning stability in every circuit or model.[^ref-fa06123d6220]
Scope of Application¶
Izhikevich's original spiking-network model linked millisecond spike-timing activity to a dopamine-sensitive state lasting a few seconds, allowing delayed reward to influence recently marked connections. Yagishita and colleagues separately stimulated glutamatergic and dopamine inputs and saw spine enlargement only when dopamine followed by 0.3–2 seconds. One is a computational credit-assignment model, the other a bounded structural experiment; their windows and endpoints are not interchangeable.[ref-fa06123d6220][ref-2560f66cc7ed-2]
Clarity¶
One delayed version has a recent pre/post event leave a local “eligible” mark. A later signal can influence marked connections while leaving others less affected. Other three-factor rules combine the same three inputs without first making that mark. When a trace is used, its duration matters.
Manages Complexity¶
The structure distinguishes connection-specific pre/post information from extrinsic modulation of the update. Some versions separate which synapses participated from whether a later outcome warrants change. The general rule remains separate from any particular neurotransmitter or circuit implementation.
Abstract Reasoning¶
Identify the local pre and post variables, the additional modulator, and their causal roles in the update. If the modulator is delayed, ask how local eligibility persists. A two-factor STDP rule with no distinct modulation does not pass this test; neither does a reward signal that has no connection to local synaptic activity.
Knowledge Transfer¶
The broader pattern is a local update shaped by an additional contextual variable; candidate marking followed by later gating is one version. The specific abstraction remains a synaptic plasticity rule; generic reward learning can use other mechanisms.
[^ref-325874ea5d3c]: Frémaux and Gerstner, “Neuromodulated Spike-Timing-Dependent Plasticity, and Theory of Three-Factor Learning Rules” (2016). [^ref-fa06123d6220]: Izhikevich, “Solving the Distal Reward Problem through Linkage of STDP and Dopamine Signaling” (2007), author-hosted original article summary. [^ref-2560f66cc7ed]: Yagishita et al., “A critical time window for dopamine actions on the structural plasticity of dendritic spines,” Science (2014), original-study abstract; full discussion not inspected in this review. DOI: 10.1126/science.1255514. [^ref-2560f66cc7ed-2]: Yagishita et al., PubMed original-study abstract, separately stimulated glutamatergic/dopaminergic inputs and 0.3–2-second spine-enlargement window. DOI: 10.1126/science.1255514.
Neighborhood in Abstraction Space¶
Three-Factor Learning sits in a sparse region of the domain-specific corpus (79th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Neuronal Signaling & Plasticity (14 abstractions)
Nearest neighbors
- Synaptic Plasticity — 0.85
- Neurotransmission — 0.83
- Spike-Timing-Dependent Plasticity — 0.83
- Corollary discharge theory — 0.82
- Retrograde Signaling — 0.81
Computed from structural-signature embeddings · 2026-10-08