I''m lookin for suggestions for a small experimental study in cognitive psychology, something that can be carried out in a couple of weeks, doesn''t require too many participants and is somehow interesting to AI people. Something about learning, memory or perception. Pattern recognition maybe. Any ideas?

why not do domething with neural networks and associative memory.
It is a fairly easy subject to read up on, find info about and IMHO has lots of connection to cognitive science...



--Spencer

"Relax, this dragon is sleeping..."

Not likely that you could do that in a couple of weeks Spencer.

LNK2001, one of my colleages is a neuropsychologist. She''s on the phone at this time, but I''ll have a chat to her this afternoon and see if we can come up with an idea for your experiment. I''ll get back to you shortly.

Cheers,

Timkin

Thanks, suggestions are welcome

bump

Okay, we've come up with a really good idea! This experiment tests visual search and focus of attention skills, which are particularly important in AI for agents that need to be able to identify items in a cluttered field of view and track them. For example, video surveillance systems that track people, assembly line robots that identify faulty parts on a conveyor belt, simulated agents in games that are looking for a specific enemy in a group, etc.

The task requires the partipant to find and identify a symbol within a cluttered field of view and report its position. The easiest symbols to use would probably be characters from the Latin (Roman) alphabet (i.e., a,b,c,d,etc.). These can be embedded in a field of other characters (characters from other alphabets (Greek, Arabic, etc) and other non-alphabet symbols). There should be three tasks:

1: Find the coloured symbol in a field of black symbols (on white), or white on black.
2) Find the Latin character in a field of non-Latin characters.
3) Find a specific Latin character in a field of other Latin characters.

In each of these tasks, there should be three sub-tasks:

a) Characters are aligned in rows
b) Characters are aligned in concentric circles
c) Characters are scattered randomly over the page (at varying angles)

For each task, the respondant needs to report the location of the desired character: just use a simply number-letter grid, as in a street directory. They should be required to report the grid reference letter first, then number. For example, B6.

Record the answer given (for correctness) and the time taken to respond.

You can then evaluate the difficulty of different tasks and sub-tasks based on the response times for correct vs incorrect answers, both intra- and inter-task.

You can also add an inhibition test into the task. If the respondant identifies a given character, say the letter X, within a radius of grid squares (perhaps 3 squares in any direction) then they should reverse the location report of the original character. So, in the example above, they would report 6B, rather than B6, IF they say the letter X within a given radius.

The idea of this is that you are trying to see if there is an inhibition effect, which would be determined if there was a significant delay in the response time in such tasks over other non-inhibited tasks. You can learn more on inhibiting cognitive functions from the literature, or by talking to your cog-sci professors.

If you need more specific details about this idea, just holler.

One caveat though: I'd appreciate being given a copy of the data you collect once you've finished. You can reserve the right to publish the results, but I'd still like access to the data in case I, or my colleague, want to do further research in this area.

Thanks,

Timkin

[edited by - Timkin on September 10, 2003 12:28:23 AM]

Put a couple of students in a maze, change some doors and see if they take the shortest route or something ^,^