= Eligibility trace
{title2=$e_t=\gamma\lambda e_{t-1}+x(s_t)$}
A decaying record of recent activity that marks a state or connection as eligible for modification by a later teaching signal. An accumulating value-learning trace is $e_t=\gamma\lambda e_{t-1}+x(s_t)$, giving $\Delta w=\alpha\delta_t e_t$. Other traces record action-policy <gradients> or pre/post synaptic events. The update and episode-boundary convention must be specified; an <eligibility trace> is not automatically identical to the visual activity trace used for <trace learning>.
Back to article page
