Testing the RDM with different 'short-term memory' decaying rates.

<p>Results are displayed as the difference in the probability of finding output biased choices produced with a 'Gellerman-like' (GLS) minus a 'Full Random' (FRS) training schedule, after being processed by the 'Rational Decision Maker' (RDM) algorithm (<i>i</i>.<i>e</i>. <i>ΔP = P</i>_{<i>output_bias[GLS]</i>} − <i>P</i>_{<i>output_bias[FRS]</i>}). Panels are arranged in columns (<i>k</i>₁ = 0.5, 0.75 and 0.90) and rows (<i>k</i>₂ = 0.5, 0.75 and 0.90). Increasing <i>k</i> values implies longer short-term memory representatioτns. <i>k</i> values of 0.5, 0.75 and 0.90 correspond to a <i>τ</i> of 1.44, 3.47 and 9.49, respectively.