Abstract
The selective attention for identification model (SAIM) is an established model of selective visual attention. SAIM implements translation-invariant object recognition, in scenes with multiple objects, using the parallel distributed processing (PDP) paradigm. Here, we show that SAIM can be formulated as Bayesian inference. Crucially, SAIM uses excitatory feedback to combine top-down information (i.e. object knowledge) with bottom-up sensory information. By contrast, predictive coding implementations of Bayesian inference use inhibitory feedback. By formulating SAIM as a predictive coding scheme, we created a new version of SAIM that uses inhibitory feedback. Simulation studies showed that both types of architectures can reproduce the response time costs induced by multiple objects—as found in visual search experiments. However, due to the different nature of the feedback, the two SAIM schemes make distinct predictions about the motifs of microcircuits mediating the effects of top-down afferents. We discuss empirical (neuroimaging) methods to test the predictions of the two inference architectures.
Highlights
In 2003, Heinke & Humphreys [1] introduced the selective attention for identification model (SAIM) to model translation-invariant object identification in multiple object scenes
Heinke and Humphreys demonstrated that SAIM could explain a broad range of empirical phenomena typically associated with selective visual attention, such as the effects of spatial cuing, object-based selection and the response time costs of recognizing multiple objects
We first performed validation simulations to ensure excitatory matching (EM)-SAIM can replicate the simulations of multiple object cost in terms of reaction times, as reported in Study 2 of Heinke & Humphreys [1]
Summary
In 2003, Heinke & Humphreys [1] introduced the selective attention for identification model (SAIM) to model translation-invariant object identification in multiple object scenes. A foundational assumption of SAIM is that the brain implements a soft constraint satisfaction as implemented by the parallel distributed processing (PDP) paradigm [2] This led to a neural network architecture with feedback loops that enable an interaction between top-down information (i.e. knowledge about objects stored in an object identification stage) and bottom-up information (i.e. sensory information). The common theme here is a dynamical implementation of a universal prior in object recognition; namely, that only one object (i.e. the winning or selected hypothesis) can cause sensory input at any one time This fundamental prior is generally mediated by lateral interactions in neuronal schemes. These simulations show that EM-SAIM reproduces the well-known multiple object cost; i.e. the increased time it takes to detect a target object with increasing numbers of non-target objects This ubiquitous empirical finding is an emergent property of SAIM’s WTA mechanism. We conclude this section by demonstrating that EM-SAIM can reproduce multiple object costs
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
