1 / 1

A few pilots to drive this research Well-trained subjects: 15 hours, including 5 of practice. - PowerPoint PPT Presentation

  • Uploaded on

. *. Duration of the primer on RT to say "Same". Complexity of decision on RT to say "Same". Number of differences on RT to say "Different ". Complexity of decision on RT to say "Same". Number of features. Vision. To join the authors: Denis.Cousineau@ umontreal.ca. Plan p Plan p ´ q

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
Download Presentation

PowerPoint Slideshow about 'A few pilots to drive this research Well-trained subjects: 15 hours, including 5 of practice.' - ivrit

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
A few pilots to drive this research well trained subjects 15 hours including 5 of practice


Duration of the primeron RT to say "Same"

Complexity of decisionon RT to say "Same"

Number of differenceson RT to say "Different"

Complexity of decisionon RT to say "Same"

Number of features


To join the authors:Denis.Cousineau@umontreal.ca

  • Plan p

  • Plan p ´ q

  • Plan (p)



"Same"-"Different", cue validity and detection task fitted by a parallel race model: The ubiquitous presence of primingDenis Cousineau, Sébastien Hélie, Christine Lefebvre, Université de Montréal

The signature of priming

Priming seems to have a typical signature, seen in the data as a concave curve which is a function of complexity (A & D below), duration (B) and number of cues (C). This pattern of results is seen in simple tasks having similar experimental procedures.

A model to model redundancy

  • Since weighted connections cannot accommodate the various results, we set all connection weights to 1.

  • We explored redundancy. A single piece of evidence can travel through a large number of redundant channels.

  • Clearer, stronger and longer signals activate a larger number of detectorsr, the number of active channels, is a linear function of the "clarity" of the signal

  • More difficult responses, resulting from more complex stimuli, requires higher thresholds from the deciders k, the size of the accumulator, is a linear function of the complexity of the signal

  • All the channels are racing to fill a decider and all the deciders are racing to make a response{this is a parallel race model, Cousineau, Goodman and Shiffrin, in press}

A few pilots to drive this research

  • Well-trained subjects:

    15 hours, including 5 of practice.

  • Stimuli:



  • Task:

    "Same"-"Different" task a la Bamber (1969)

    Pilot 1: Duration of primer and complexity of decision with holistic stimuli.

    Pilot 2: Complex decision with separable stimuli.

  • "Different" responses suggest a serial self-terminating search for the first difference BUT!

  • "Same" responses are concave and faster than "Different", rejecting any serial model (Sternberg,1998).

A) The "Same"-"Different" task

Bamber, 1969

What is priming?

Priming is a phenomenon in which facilitation occurs in low-level tasks. Mechanisms of priming may also play a central role in other phenomenon. See "The signature of priming" section.

What is the cause of priming?

Priming might be some reminiscent activation left in the system after the presentation of a primer. Assuming a thresholded network of connections, predictions can be derived. See the section "A model to model redundancy".

What factors modulate priming?

Stronger, clearer and longer signals generate reminiscent activation in a larger number of channels. This assumes that the channels are highly redundant. However, more complex decision can result in a more stringent threshold. See the section "Predicting priming".

Preliminary tests

We explored the impact of duration of the primer and complexity of the decision in a "Same"-"Different" task. The signature of priming was found. Using holistic or discrete objects changed the results, as predicted. See "A few pilots to drive this research".

What is priming then?

Priming may have a highly adaptive value: parallel systems operating in real time must be able to anticipate the processes to come next in order to reduce the number of possibilities. Thus, internal priming is the most natural outcome of PDP.

Probe 


  • The probe complexity C (string length) was 1 to 4;

  • Duration of the first slide not controlled by Bamber;

  • If different, the probe had from 1 to C differences.

Duration has a concave effect (reproducing Arguin and Bub). This suggests that the number of reminiscent activations (r) is the only factor changing with duration.

Complexity of holistic stimuli has a linear effect. This suggests that the subjects are increasing their threshold with increased complexity of the object (as well as receiving less evidences r).


B) The "letter"-"non-letter" priming task

  • With no primer (neutral), there is no effect of the duration D.

  • With a primer, responses are concave and faster than neutral conditions.

  • The fact that the results and the experimental procedures in A & B are identical suggests that the same mechanisms are active.

Arguin & Bub, 1995




  • The complexity C of the probe is always 1

  • The duration of the prime D is varied (50..200 ms).

To say different, there is no interaction of complexity of the objects with the number of differences between the primer and the probe. This suggest a constant threshold to say "Different".

Predicting priming

  • Altering the "clarity" of the primer (such as its duration) will leave a larger number of reminiscent channels which are easier to reactivate. Reducing r alone predicts a concave curve.

  • Increasing the complexity of the input will necessitate a larger k. However, the activated channels will be spread out and more likely to decay. Reducing r and increasing k predicts a straight line.

  • Curiously, if we could change k while keeping the number of activated channels constant, we would inverse the curve. Increasing k alone predicts a convex curve.

C) Number of masks in a cued detection task

  • For a given cue validity, the decrease in accuracy is larger between 4 and 8 masks than between 1 and 4.

  • Strength theories cannot accommodate these results, including weighted neural network.

1. Concave:r alone changes

2. Straight:r and k changes

3. Convex:k alone changes

Shiu & Pashler, 1997



  • Complexity C and duration D are held constant at 1 and 50 ms resp.

  • The number of masked locations following the probe is varied (1, 4 or 8).

Here, complexity has a concave effect (reproducing Bamber). This suggests that the threshold operates on individual letter and is not affected by the length of the string.

To reject a whole string, there is an interaction of complexity with the number of differing letters. The threshold in this case seems to be modified by the complexity of the string to reject.

D) A feature detection task

  • The small decrement in accuracy when increasing the number of features (complexity) from 1 to 2 compared to the large decrement between 3 and 4 is against predictions of limited-capacity models.


Cousineau & Shiffrin, in prep.



  • Detecting well-learned features/configurations is easier;

  • There is no primer (Ss were trained in a different task), suggesting that preactivation can be internalized.