Week 3 Flashcards
Template theory
Recognition of an object follows a ‘match’ between the stimulus and an internal construct (template) of that object
Template theory issues
- Assumes an exact match
- Requires a huge number of templates for all variations of an object
- We can recognize abstract objects
Feature theory
All objects are composed of separable distinct parts - features
- Like a visual alphabet
Feature theory support
- Neuronal responding to specific things (edge detection, movement detection, colour detection, etc.)
- Cells respond to a very specific simple feature
- Visual search tasks (multi-feature takes longer)
Cognitive visual searches
Press button when you see…
Conjunction search
Look for multiple features of an object in a set (ex. Red and horizontal lines)
Search asymmetries
Easier to find the “tilt” than “not tilt” indicating the feature we use is “tilt” not “absence of tilt”
- Consistent with brain physiology
Pop-out effect
Something is immediately recognizable as different in a set
- happens regardless of number of things in a visual field
Structural theory / recognition by components
- Objects are created by combining geons by their edges
- Like an object alphabet with 36 distinct geons
- Geons can be recognized from any angle
Issues with RBC theory
- Some things require very specific idiosyncratic geons (specific) - but they’re supposed to be universal not specific
- Ignores context
- Doesn’t explain differences within object type (ex. apples vs oranges)
Recognition by multiple views
- Template theory but we have templates for multiple angles
- Viewpoint dependent
Face processing
- Faces are processed specially
- Holistically processed, not feature based
Face inversion effect
- Upside down faces hard to recognize than other upside-down objects
- Faces usually processed holistically, but must be processed by parts when upside-down
- Monkeys don’t have this issue bc they’re upside-down often
Prosopagnosia
Can’t recognize faces
- Either due to brain damage or just a thing from birth
Super-recognizers
- Can recognize faces super well, even if the picture is degraded (from and odd angle or distance), they haven’t seen them for a while, can tell easily if two faces are from the same person or not
Better eyewitnesses, awkward if they know you but you don’t know them
Tachistoscope
An apparatus for the brief exposure of visual stimuli that is used in the study of learning, attention, and perception
ie. pre-computers
Recognition threshold
Time required to recognize a word
Recognition tasks
- Used to do with a tachistoscope
- Stimulus presentation followed by a post-stimulus mask (ex. Random shapes or letters covering it)
- Recognition depends on familiarity
Ex. more frequent words recognized more often when masking task done - Also depends on priming
Chronometric approach
Assume mental processes take time, research how long they take
5 things that affect word recognition
- Frequency
- Repetition priming
- Context effects
- Well-formedness
- Overregularization errors
Frequency
- People are faster to recognize common words, doesn’t mean we don’t know the uncommon ones
ex. Happy vs. hippy or Happy vs. harpy
- Word frequency makes it faster to recognize NOT familiarity
Repetition priming
- Viewing a word once helps you read it next time
- Priming is like a brain warm-up
Context effects
- Helps us recognize objects and can change what we perceive
- Top down processing
Well formedness
- word superiority effect works with well formed words, even if they’re not real words (ex. LAFE)
Why? Guided by our knowledge of spelling patterns + our bigram detectors
Word superiority effect
It is easier to perceive a letter when it is part of a word than when it is presented on its own. (ex. of a context effect)
- not just educated guessing
Bigram detectors
Detect letter pairs
- in word superiority effect
- Goes with priming, more common pairings expected
Overregularization errors
- Making errors
- Will misread less common words as more common ones
Patient DF
Drawings
- Had brain damage in lateral occipital cortex (LOC) that recognizes objects
- Drew simple disconnected lines instead of the actual items - when she was looking at them
- Drew more accurate to real life - when drawing from memory
Example of apperceptive agnosia: can see but can’t organize objects to perceive an entire object
Patient Dr. P
Associative agnosia
- Can see but can’t link what they are seeing to basic visual knowledge
- Thought a glove was a pouch or purse
- Mistook his wife’s head for a hat
Bottom-up vs. top-down
Top down processing → concept driven
Bottom up processing → data driven
Knowledge representations
Local representation → storage in a particular location (ex. preferential response in one cell)
Distributed representation → knowledge represented by a pattern of activations distributed across a network
ex. bigram detectors
Reading fonts
- Harder to read fonts seen as written by a less intelligent author
- It’s more difficult to read words in all caps
Why?
All lines the same height and none go below/above (like in lowercase j)
Feature net
- How to build a system that recognizes words
- System of feature detectors wired together
- Neurons light up and are connected to more neurons
- Parallel processing models in neural networks
- Have a good idea of what features make up letters and words
- Each detector has an activation level
Activation of feature detectors
- Each detector has a particular activation level = how energized it is
- Will eventually reach the detector’s response threshold = it will fire
Are NOT neurons
Activation level
- How activated something is
- Detectors that have recently fired (warm-up effect) or that have fired frequently (exercise effect) will have a higher activation level
Efficiency vs. accuracy
Effect of inferences
Inferences cause a slight decrease in accuracy but a huge benefit in efficiency
McClelland and Rumelhart Model
- Pattern recognition
- Excites connections for specific words that inhibits connections for others
- Excitement: activation of one node/detector activates another due to their link
Ex. TRIP inhibits TRAP and TAKE
Speed reading
- Due to efficient and accurate inferences
- Read by inference rather than word-by-word
- Get faster by moving your finger faster along a line of text, then follow that instead of letting the finger follow you
- Basically just guessing more than before
- Easiest for non-technical readings