[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: historical perspective on theories of attention

Dear Diana & All,

In terms of considering using rapid switching as the basis of selective attention, I did do some work in developing control algorithms based on this principle. The basic idea was to switch between negative feedback and positive feedback, depending upon some control criteria. Without getting into details, the dynamic response of a simple controller was most interesting. First of all, when the controller achieved zero error, the switching rate tended towards the maximum limit (infinity in theory). Second, the response to an error, showed a trajectory where there was no overshoot as normally occurring using algorithms based on linear theory. In addition, the response provided a substantially smaller integrated absolute error as compared to an equivalent controller based on linear theory. A casual observation of our motor movement supports this response in that no overshoot is usually observed.

In the controller design, the control criteria for switching was based on both the error and its rate of change, basically bringing us into the phase domain of analysis rather than methods based on Fourier. It is permissible to use the auditory stimulus as an error, leaving us with the question as to what the equivalent control criteria should be to perform some kind of frequency analysis rather than control, as there seems to be no getting away from this fundamental requirement. Determining F0 in the case of a single source or multiple F0's in case of multiple sources on a continuous basis (the F0's could be changing) becomes critical for source segregation, leaving for us to chose the basis by which a particular source is selected.

Ignoring for the moment how general arousal changes (as while sleeping), it seems that selection of the source depends critically upon the rate of change, whether it be frequency, energy level or results from binaural processing. If rate of change is used as the basis of selection rather than "level" as used in your paper, a clearer criteria becomes available to model such behavior and it could be applied to every level of hierarchical structures. It also becomes easier to explain the need for hierarchical models of the neural system.

All this is of course contingent on selecting the appropriate control criteria and I have chosen evaluative bivalence, a concept well described in the field of psychology, but used by me for signal processing. I have been documenting my progress at my website www.tonepitch.com and hope to update it soon with some results showing how speaker independent vowel identification can take place. At the moment, spectral centroid provides no relevant information other than speaker dependency.

I am sorry to have made this email so long and hope that it is relevant to the present discussion.
Cheers to all,

Randy Randhawa

Diana Deutsch wrote:
Dear Dan, John et al.,

To place this discussion in historical perspective, the 'late selection' model of attention was first proposed by Deutsch, J.A. and Deutsch, D.,'Attention: Some theoretical considerations',/ Psychological Review/, 1963, 70, 80-90. There have, of course, been a large number of elaborations of this basic model. The article is attached.


Diana Deutsch

This note was intended to be sent to the entire list (my change of e-mail address prevented it from appearing); Dan has replied to me privately.


What you've suggested so succinctly seems a lot like the concepts on perception that I discuss on my Web site. Using a basic system analysis approach I've tried to find fundamental principles from which to establish generalized requirements for sensory systems based on what it takes for animals to survive in their environments.

Here is some of my thinking: Survival requires an animal's sensory organs to produce a timely response to environmental information. These responses must identify the relative importance of sources in a way that describes a "situation" whereby "awareness" characterizes the auditory scene. In any situation, sensors should be able to select for attention instantaneously the single most important source within the scene. To do this the model I propose has a hierarchy of perceptual levels each of which has its own ability for awareness and attention but is subservient to judgments by higher authority. Each level's responses are based upon specific time frames within which information from each level can be made available. (Simple meanings occupy less time to absorb than complicated meanings,) In addition, each level is capable of selecting a priority source within its domain for attention. For example, the lowest levels which have the simplest information could respond reflexively to a source within milliseconds. But its response could be moderated by a top-down decision from higher levels. It is thus possible, as you suggest, that what seems like simultaneous multiple-source attention at the conscious level is actually the rapid switching of priorities among source objects based on subliminal decisions at the lower levels. As an example, consider the complexity of the attention decisions a quarterback must process within the few seconds he has during a football play.

A crucial question: how to achieve and synchronize timely responses at all levels. It appears to me that this problem has not been seriously addressed in the current paradigm.

John Bates

At 12:18 PM 4/28/06, you wrote:

I think another factor to consider in the concurrent segregation of sounds is that much of the segregation may be accomplished pre-attentively. Low-level (in the brain, and even cochlea) feature detectors may segregate aspects of sounds, if not the sounds themselves, well before they percolate up to what we call "consciousness." This would depend on the type of segregating cue under consideration, whether it is pitch, spatial location, onset time, etc.

Professor Diana Deutsch
Department of Psychology University of California, San Diego
9500 Gilman Dr. #0109 La Jolla, CA 92093-0109, USA

858-453-1558 (tel)
858-453-4763 (fax)