03.+Perception

February 13th, 2015
 * 03. Perception**

Sensation & Perception Objects & Geons Concrete & abstract representations Words & Feature nets Faces
 * Outline**

CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) D efinitions:
 * Sensation & Perception**
 * Works because humans are much better at object recognition than are computers
 * Sensation: bringing information in through the senses
 * Perception: interpreting/making sense of information gathered from the senses

Recognition by Components
 * Objects and Geons**
 * We have an "object alphabet" consisting of 36 geons (geometrical ions)
 * Geons can be combined to create just about any object
 * Explains recognition of "complex" objects
 * Geons are supposed to be viewpoint independent, meaning it doesn't matter what angle you look at them from
 * Evidence:
 * We're much more likely to recognize an object if we can recognize the geon
 * We have trouble if geon intersections are removed

So where are we now?
 * Problems
 * Evidence suggests that object recognition is NOT viewpoint independent[[image:airplane.png width="191" height="112" align="right"]]
 * Evidence suggesting that object recognition is viewpoint //dependent//
 * Objects (e.g., airplanes) are easier to identify from some angles than others
 * There are brain cells that fire for:
 * a specific object
 * a particular view of the object
 * Use a mixture of theories
 * RBC (geon) theory can be used to make general distinctions between objects
 * Memory of objects (roughly 40 views for the average object) allows for finer distinctions

Big Picture Music Perception
 * Concrete & Abstract Representations**
 * Geons are __abstract__ - they can describe virtually any object
 * Viewpoint dependent memories are __concrete__ (sort of) - they describe a specific object
 * We recognize objects based on both - combining abstract and concrete is a theme in our cognitive systems
 * E.g., in language we learn rules (sing-sang) and specific exceptions (bring-brought)
 * A musical interval is the distance between two notes
 * An interval is a basic building block of music, like a geon is to object recognition, or like the alphabet is the building block of words
 * Smith et al. (1994)
 * Task: participants were asked to identify musical (minor third, major third, perfect fourth)
 * 2 Conditions
 * Extensive training group was given extensive training with the three intervals using abstract labels (this is what a minor third sounds like...)
 * Folk song label group was given examples of folk songs with those intervals (Greensleves = minor third, Kumbaya = major third, Here comes the bride = perfect fourth)
 * Results: The folk song label group did almost as well as musical experts. The extensive training group's performance was almost random.
 * Intervals are like a "musical alphabet" but we struggle to learn it.
 * It seems we memorize every song we ever learn (concretely rather than abstractly)
 * Smith et al. (1994)
 * Twinkle twinkle, the alphabet song, and Ba ba black sheep all have the same melody, but we tend not to notice because we memorize each song separately


 * Words & Feature Nets**
 * Feature nets: networks
 * Network: a collection of "nodes" that are connected to each other
 * Each feature in the network has these properties:
 * Activation: each feature in the network can be activated to varying degrees
 * Threshold: if a feature's activation surpasses a threshold, that feature "fires"
 * Influences on activation:
 * Recency: if a feature has been activated recently, it's easier to activate (e.g., repetition priming)
 * Frequency: if a feature is activated frequently, it's easier to activate (greater resting activation)
 * Example: reading (see picture)
 * Feature net with feature detectors, word letter detectors, and word detectors[[image:wordfeaturenet.png width="282" height="134" align="right"]]
 * If you see the left-bottom item, it receives activation.
 * If it's above threshold, it sends activation to the letter C through its connection
 * If C receives enough activation, it fires
 * C is more likely to fire you've seen C frequently or recently
 * Note: features act like neurons, but they are not neurons
 * McClelland & Rumelhart model: Inhibitory connections, not just excitatory
 * Connections both ways: bottom-up and top-down

Is face memory special?
 * Faces**
 * Maybe yes:
 * We are really good with upright faces, but not so good when they're upside down
 * Associated with fusiform gyrus - damage selectively impairs face recognition (prosopagnosia), not object recognition. This implies objects and faces use different systems.
 * Maybe no...
 * It applies to highly familiar stimuli, not just faces
 * Dog show judges are good with dogs
 * Birders are good with birds
 * Experience matters!