Lecture 3: Pattern Recognition & Attention Flashcards
What are template models?
- Match stimulus to template in memory
- Support for idea: computers/ scantrons
- Limits: inefficient, irregular world, strict match (lead people to go to something a little more simple, feature models)
What are feature models?
- Features are more regular than patterns
- Complex objects composed of simple features
- Do we process patterns as collections of features?
What supporting evidence did Neisser (1964) find for feature models?
- High speed scanning (this is behavioural evidence for feature processing)
- When people look at patterns they extract those Features and the system makes notes of those features and then their compared to some type of representation that we have stored
- Compared to “z” features
- When the letter z is with other letters that don’t share features, people find it more quickly. When it is with letters that do share features, it is harder and takes much longer
What supporting evidence did Lettvin et al. (1959) find for feature models?
- Microelectrodes into cells of frog retina
- Recorded activity: Edge detectors, Moving edge, Dimming, Convex edge (small, circular dot moves)
- All these cells process certain types of features, tells the frog that a fly is going by.
- Shows the existence of feature detectors
What supporting evidence did Hubel & Weisel find for feature models?
- Simple cells (these cells only respond to): Simple patterns of light, Location specific, Edge, slit, line
- Complex: Same as simple, but not location specific
- Hypercomplex: Moving lines
- W,x,y: Movement speed
What is beyond features?
- We also have Top-down (conceptually driven) pattern recognition
- Gestalt grouping principles: processing of “whole” object
- Neuroimaging evidence shows specialize brain areas for processing of whole visual objects
• Pattern recognition is influenced by knowledge (random mix of letters from the word university vs. the word university)
What was the study by Avant & Lydall on masking known words vs. fake words?
- Masking of Boy vs. YOB
- Results: a shorter interval is required to erase (mask) BOY than YOB (because we recognize boy as a pattern and have knowledge of it, we have to shorten the interval and get at it quicker in order to erase it)
How does reading illustrate top down processing?
- Average of 7,500 features per page (100 per sec)
- Role of top-down? not feature-by-feature
What is the word superiority effect?
- Reicher (1969),Wheeler (1970)
- They showed people a single letter, and masked it. After they masked it, they ask you whether the letter was a D or G. The accuracy is about as low as chance if the interval is at the right time. Then they put the same letter “D” but at the end of the word, “wind”, and then they masked it and asked them if they saw the word wind or wing.Not much has changed except it is now a letter at the end of a word instead of a letter on its own. Even though there are more features to process here, they actually perform much better (they are really accurate).
- Performance better with wind than D
- Top down influences pattern recognition
What is repetition blindness?
- Use of RVSP (rapid serial visual) paradigm (Morris and Harris, 2002)
- Show people a series of items in the same spot rapidly
- “when she spilled the ink there was ink all over”
- People fail to see stimulus (“ink”) the second time
- Top down influence: cognitive system has identified the stimulus, so “expect” NOT to see it again (system has already processed it and is inhibiting us from seeing it a second time)
- “I broke a wine class in my class yesterday”. People read “wine glass” (top down context biases)
- Also, expect repetition blindness to class- class, but there was no repetition blindness to the 2’nd class: top down (“glass”) overrides effect (“glass”-“class”)
- Tells us that Top down influences pattern recognition (word knowledge effects pattern processing)
What is the RBC theory?
- Recognition by components theory (Beiderman, 1987, 1990)
- Objects made up of combinations of “geons”
- Recognition involves:
- 1’st parse objects into component geons
- note where geons join (find edges)
- match geon combinations to representations in memory
- very bottom-up model
What are the problems with the RBC theory?
- Expertise and experience affects early perception of objects
- Overall (whole object) can be perceived as fast as components- not just bottom up
What is agnosia?
- Greek: Prefix “a” meaning “not” and “gnostic” meaning to “to know”
- Agnosia is a failure or deficit in recognizing objects:
- Pattern of features cannot be synthesized into a whole
- Or person cannot connect the whole pattern to a meaning
- Caused by specific brain damage
What is prosopagnosia?
- Disruption of face recognition
- But these patients typically are able to recognize other objects
- Man who mistook his wife for a hat
What is apperceptive agnosia?
- Disruption in perceiving whole patterns
- Can process basic features (lines, Colours)
- Cannot integrate into a whole object
- Typically have damage in the right hemisphere parietal area
What is associative agnosia?
- can combine features into a whole, can copy or describe a drawing
- cannot associate with a meaning (cannot identify)
- involves temporal lobes of both hemispheres
What do Agnosia studies tell us?
Recognition of objects is not “immediate”, but has 3 steps:
- Perception of features
- Integration of features into a larger whole pattern (Gestalt)
- Association of the pattern to meaning
Summary of pattern recognition
- fast
- apply to novel patterns
- across orientations, occlusions
- involves features
- augmented by top-down (conceptually driven) processes
- multi-step process: feature detection, integration, association
What is attention?
“The mental process of focusing effort on a stimulus or on a mental event”
What are the two main types of attention?
- Visual Attention
2. Selective Attention
What is visual attention?
Sensory memory has a large capacity. However, to be efficient, we must be able to select the most relevant information from the environment. How is this done?
- Space-Based Visual Attention
- Object-Based Visual Attention
What is space based attention?
- Direction of visual attention can be separated from direction of gaze (eyes). We can disassociate them (e.g., driving. Good drivers know what’s happening around them, your gaze has to be up front but you also have to know what’s happening beside you and behind you )
- Endogenous: internal/intentional (a top down process)
- Exogenous: drawn by external cue (not me moving my attention, something else moving it) (bottom up process)
What is the spotlight Metaphor (space based attention)?
- Visual attention as a spotlight
- Spotlight focused (endogenous or exogenous) on locations in space
- Spotlight facilitates processing or information at the attended location (information at unattended location does not get processed as efficiently)
- Information at unattended locations is inhibited
Why space based attention?
Exogenous cues: draw attention to potentially important
events in space
Endogenous cues: place attention according to expectancy
What is object-based attention?
- Attention is placed on objects
- within-object processing is more efficient (faster, more accurate) than across-object processing
What was Lavie and Driver’s (1996) study on object based attention?
- Line (object) study
- near and far conditions showed that it was not about the distance between the lines because they were the same distance so they concluded that it was about switching from one object to the other
- Faster to targets (dots or gaps) on same “object” (line)
- Object based attention
Why object based attention?
- major goal of vision is to identify objects
What is the filter theory (Broadbent, 1958)
3 components:
- selective filter
- limited capacity decision channel
- long term memory (detection)
What is the evidence for the filter theory?
- when told to recall by ear (left ear: 4, 7, 3 vs. Right ear: 1, 9, 5): 65% of items. Attend to one ear then switch quickly to other.
- told to recall sequence (4-1, 7-9, 3-5: only 20% of items. Requires rapid switching of the selective filter.
- suggests select info based on channel (i.e., left channel vs. right channel)
What is selective attention?
SM has a large capacity but does not last very long. Need to quickly select information out of SM for further processing. How is this done?
- filter theory
- Attenuation theory
What was Cherry’s research on selective attention?
- Dichotic presentation: shadowing
- He Varied the content of unattended: change male/female (easy to detect) OR changed language (did not notice language change)
- suggests that we attend / select using physical cues
How did the class Demo/ Moray’s study challenge or demonstrate problems with the filter theory?
- Class demo (Girl reading in monotone/choppy voice, guy shadowing and Prof. reading magazine in same voice. Then they do it again but instead of reading the magazine in a choppy voice prof says the guys name in a bunch of sentences. The guy shadowing hears his name even though it was in his unattended ear)
- Moray (1959) person’s name on unattended channel (aka the cocktail party effect)
How did Gray & Wedderburn’s study challenge the filter theory?
- dichotic presentation (one stimulus to one ear and a different stimulus to the other ear at the same time. Car and 3, 9 and Pen, Ter and 5)
- They found that people did not organize it by channel or ear but rather responses were organized by word/digit (i.e., Carpenter 395)
- Shows: Selected by semantics
How did Corteen & Wood’s study challenge the filter theory?
- Phase 1: Presented people with little shocks (UCS) which lead to a galvanic skin response (UCR). Paired shock with names of cities (CS) and eventually they would present just the city name without a shock and people would have a galvanic skin response
- Phase 2: Dichotic listening. Old (Denver) and new city names (Toronto), Attended vs. unattended ears (according to Broadbent’s filter model if the city name comes up in the unattended ear then they should not have a GSR because they did not process it.)
- Results: GSR’s unattended=attended (equal magnitude). GSR’s to old and new names.
- Conclude: semantic info from unattended is processed (against Filter Theory)
What is the Attenuation theory (Treisman, 1964)?
- Filter: Is not all-or-none, Just attenuates, Uses physical
- Dictionary units: “lexicon” (logogens- logogens are mental representations of words in our lexicon. Logogens operate like neurons). Affects detection.
- Thresholds: Variable (name vs. house). Logogens have different thresholds, words that are significant to use have low thresholds (e.g.., your name)
