Three-factor learning
{{confusing|date=January 2025}}
In neuroscience and machine learning, three-factor learning is the combinaison of Hebbian plasticity with a third modulatory factor to stabilise and enhance synaptic learning.{{Cite journal |last1=Porr |first1=Bernd |last2=Wörgötter |first2=Florentin |date=October 2007 |title=Learning with "Relevance": Using a Third Factor to Stabilize Hebbian Learning |url=https://direct.mit.edu/neco/article/19/10/2694-2719/7218 |journal=Neural Computation |language=en |volume=19 |issue=10 |pages=2694–2719 |doi=10.1162/neco.2007.19.10.2694 |pmid=17716008 |issn=0899-7667|url-access=subscription }} This third factor can represent various signals such as reward, punishment, error, surprise, or novelty, often implemented through neuromodulators.{{Cite journal |last1=Gerstner |first1=Wulfram |last2=Lehmann |first2=Marco |last3=Liakoni |first3=Vasiliki |last4=Corneil |first4=Dane |last5=Brea |first5=Johanni |date=2018-07-31 |title=Eligibility Traces and Plasticity on Behavioral Time Scales: Experimental Support of NeoHebbian Three-Factor Learning Rules |journal=Frontiers in Neural Circuits |language=English |volume=12 |page=53 |doi=10.3389/fncir.2018.00053 |doi-access=free |issn=1662-5110 |pmc=6079224 |pmid=30108488}}
Description
Three-factor learning introduces the concept of eligibility traces, which flag synapses for potential modification pending the arrival of the third factor, and helps temporal credit assignement by bridging the gap between rapid neuronal firing and slower behavioral timescales, from which learning can be done.{{Cite journal |last1=Frémaux |first1=Nicolas |last2=Gerstner |first2=Wulfram |date=2016-01-19 |title=Neuromodulated Spike-Timing-Dependent Plasticity, and Theory of Three-Factor Learning Rules |journal=Frontiers in Neural Circuits |language=English |volume=9 |page=85 |doi=10.3389/fncir.2015.00085 |doi-access=free |issn=1662-5110 |pmc=4717313 |pmid=26834568}} Biological basis for Three-factor learning rules have been supported by experimental evidence.{{Cite journal |last1=Kuśmierz |first1=Łukasz |last2=Isomura |first2=Takuya |last3=Toyoizumi |first3=Taro |date=2017-10-01 |title=Learning with three factors: modulating Hebbian plasticity with errors |journal=Current Opinion in Neurobiology |series=Computational Neuroscience |volume=46 |pages=170–177 |doi=10.1016/j.conb.2017.08.020 |pmid=28918313 |issn=0959-4388|doi-access=free }} This approach addresses the instability of classical Hebbian learning by minimizing autocorrelation and maximizing cross-correlation between inputs.
References
{{Reflist}}