Abstract
The Integrated Reweighting Model has been used to replicate a variety of phenomena in Visual Perceptual Learning, or VPL. These phenomena include learning, location specificity (Dosher et al 2013, PNAS), and even roving (Dosher et al 2020, JOV). While the phenomena that have been replicated span many of the important characteristics of VPL, the tasks replicated have been primarily constrained to single forced-choice orientation discrimination tasks. Here we demonstrate a twin network modification to the model allows it to perform a 2-AFC (alternative forced choice) contrast discrimination task. Thus, as compared with previous versions of the model that focused primarily on orientation judgements of single images, this modification expands the model in capability both to 1) compare two separate images and to 2) make contrast judgments. The twin network methodology uses two parallel instantiations of a variant of the classic model whose outputs are then compared. This model thus offers each trail, in the case of contrast discrimination, an estimate of which image has higher contrast. Here we test this model in a classic behavioral contrast discrimination protocol inspired by Yu et al in 2004 in JOV. The similarity to classic protocols in VPL is made possible by the twin network instantiation of the integrated reweighting model. Expanding the capability of classic models is critical for determining their utility as models for the human visual system.