Figure 4.

Input variance and prior knowledge can affect the nonlinearity and the threshold of subjective visibility reports. (a) Parametrically varying stimulus strength directly changes the amplitude of the input vector and leads to a nonlinear pattern of subjective visibility reports. The slope and intercept of the resulting sigmoid depend on stimulus variance: low variance leads to an all-or-none relationship between the evidence and the visibility reports (i), whereas high variance leads to a more linear relationship as well as an increase in the visibility threshold (ii). (b) Prior knowledge can also affect the visibility threshold. Increasing the prior probability of the absent class increases the visibility threshold for all stimuli, thus lowering subjective visibility reports. When only the prior probability of X is increased (capturing ‘hysteresis’ experiments where subjects come to expect the next stimulus), then the visibility threshold is lowered for X alone, while the visibility threshold for Y barely changes. (c) Visibility and discrimination interact when both priors and stimuli variance are varied. If the probability of the absent class is relatively low (or similarly if the evidence is relatively high), increasing the variance reduces both visibility ratings and discrimination performance. However, when P(absent) is high (or similarly, if the evidence is low), increasing the variance can diminish discrimination performance while increasing visibility ratings. This diagram captures the paradoxical finding that increased attention can lead to reduced visibility [26].