Oxyopia Abstract

June 14, 2002
Noon
489 Minor Hall

Zhong-Lin Lu, PhD
Laboratory of Brain Processes (LOBES), Dept. Psychology and Neuroscience Graduate Program, University of Southern California, Los Angeles
Host: Martin Banks

Title

"External noise methods, observer models, and mechanisms of attention/perceptual learning"

Abstract

The external noise method and the linear amplifier model (LAM) have been widely used to reveal inefficiencies of the perceptual system (Pelli, 1981). However, the LAM generally cannot accommodate threshold versus external noise contrast (TVC) functions at multiple criterion levels, even though it can adequately model single TVC's (Lu & Dosher, 1999). We proposed a "triple-TVC" method - measuring TVC's at three performance levels - to measure the nonlinearities in the perceptual system in addition to additive internal noise. We also proposed a perceptual template model (PTM), an elaboration of the LAM model with a non-linear transducer function and internal multiplicative noise (or equivalently, contrast gain control), to model TVC functions at multiple performance criterion levels as well as full psychometric functions for a full range of external noise levels. The estimated parameters of the PTM are independent of performance criteria. Related to the objectives of the triple-TVC method, the double-pass method (Burgess & Colborne, 1988) tests the observers with the exact same stimulus (signal + external noise) twice (thus "double pass") to obtain response consistency as an index of the amplitude ratio of internal and external noise and therefore estimates of multiplicative noise when internal noise is not constant. We review the mathematical properties of four observer models, the LAM, the LAM with decision uncertainty (Pelli, 1985), the multiplicative noise model (Burgess & Colborne, 1988), the multiplicative noise + uncertainty, model (Eckstein, Ahumada & Watson, 1997)), and the PTM (Lu & Dosher, 1999) and their ability to account for results from both the triple-TVC and double-pass paradigms. We conclude:

• LAMs can accommodate neither TVC's at multiple performance criterion levels or the consistency data;
• LAMs with uncertainty and a max decision rule could accommodate TVC's at multiple performance criterion levels; they can't accommodate double-pass results;
• Both PTMs and EAWs (with large amount of uncertainty and a max decision rule) could accommodate results from both triple-TVC and double-pass paradigms.

We conclude that both multiplicative noise and some form of nonlinearity (either in terms of uncertainty or a nonlinear transducer function) are necessary to model the results from the triple-TVC and double-pass experiments. The triple-TVC and double-pass methods provide complementary constrains on observer models.

The PTM model provides a theoretical framework to distinguish three mechanisms underlying performance improvements in visual attention (Lu & Dosher, 1998) and perceptual learning (Dosher & Lu, 1998): stimulus enhancement, external noise exclusion via template retuning, and reduction reduction of contrast gain-control or multiplicative noise. This can be achieved via systematic measurements of human performance as a function of both the amount of external noise added to the signal stimulus and the attention/training manipulation. A key prediction of the framework is that it is possible to isolate the three mechanisms, e.g., a decoupling of attention/perceptual learning effects in high and low noise. Mathematically, we cann use measurements of TVC functions at multiple performance levels to characterize the nonlinearities in the perceptual system and to distinguish mechanism mixtures. In visual attention, pure cases of template retuning (Dosher & Lu, 2000; Lu & Dosher, 2000) and stimulus enhancement (Lu & Dosher, 1998; Lu, Liu & Dosher, 2000) have been documented separately and in different circumstances. In perceptual learning, a mixture of stimulus enhancement and template retuning was found in visual periphery (Lu & Dosher, 1998, 1999). A pure mechanism of template retuning was found in fovea (Lu & Dosher, 2001). Implications on using the various external noise methods in conjunction with attention and/or perceptual learning manipulations to characterize mechanisms of attention and perceptual learning will be discussed.

[Back to Lu - Oxyopia Page]