Exam 1 Flashcards
Is the distinction between the mind and brain Cartesian dualism? Why or why not?
NO! Descartes thought there were two kinds of things in the universe: matter and energy. In humans, there’s a third thing: a soul/mind, which is fundamentally separate from the body and does not obey the laws of the natural world.
ACTUALLY, minds are subject to natural laws, but different ones from brains.
Characterize the distinction between minds and brains.
The mind is not equivalent to the brain. It derives from the activity of the brain, but the laws that govern the way the brain works (neurophysiology) are not the laws that govern the way the mind works (psychology)
Describe Marr’s Computational Theory level of analysis.
Computational theory level:
~Most abstract
~Essentially understanding the problem
- What problems is the system solving, and why?
- What are the constraints on its solution? (note: constraints make a solution possible)
Describe Marr’s Representational Algorithm level of analysis.
Representational Algorithm level:
~What info does the system represent about the problem and how?
~What does it do with that information? (this is the algorithm)
~What is the output, etc.
Describe Marr’s Physical Implementation level of analysis.
Physical implementation:
~How are these representations and algorithms realized in the physical hardware of the device itself?
-the algorithm is the same, even when the hardware changes
Describe the interaction of constraints among Marr’s levels of analysis.
The levels are separate, but not fully independent. For example, something you find out about one level constrains questions you can ask about another.
Each level constrains others but it is more so in the top-down direction than bottom-up.
Some kinds of hardware are better at implementing some kinds of algorithms than others.
What are Marr’s levels 4 hoomanz?
1) The world that supplies the constraints that allow a function to be computed
2) The mind!
3) The brain
Generally, what are we talking about in Marr’s paper?
~Operation of ganglion cells in the retina
- send outputs along optic nerve (axon of ganglion cell)
- photoreceptors respond to (represent) luminance at each location in the image
- ganglion cells respond to (represent) contrast
What one main goal of early vision discussed in Marr’s paper? What is the problem with reaching this goal?
GOAL: surface boundaries are useful indicators of object boundaries, so we wish to find them!
PROBLEM: retinal image tells you about the luminance at each point, not surface boundaries :c
What are the constraints that make finding surface boundaries possible in Marr’s paper? Given these constraints, what is the new goal?
CONSTRAINTS:
~Smoothness: reflectance tends to change smoothly within surfaces and abruptly between them
~Projective geometry: adjacent points in the world project to adjacent points in the image
–>adjacent points in the image tend to be adjacent IRL because light travels in straight lines
NEW GOAL: find contrast!
What is the representation and algorithm used by the visual system to find contrast?
Representation: input = luminance map (retinal image) –> output: contrast map
Algorithm: compute contrast map by convolving luminance map with a difference of Gaussians at multiple locations
What does a nonlinear change in luminance correspond to?
Surface boundary!
What is the significance of the first and second derivative of a luminance map?
The first derivative shows where the slope of the luminance map changes – if that change is abrupt, there will be a corresponding dip in the first derivative graph.
The second derivative shows where the slope of the first derivative changes – if there is a dip, there will be a corresponding zero crossing in the second derivative graph.
What is a zero crossing? What is its significance?
A zero crossing is negative on one side, and positive on the other. This indicates an edge location.
How do you compute a difference of Gaussians?
Point-by-point, subtract one Gaussian from the other. We are left with a Mexican hat distribution
What is the algorithm used to detect contrast?
Convolution of DOG with luminance map.
basically, take the dot product of the vectors that represent the DOG’s height and the luminance of the image
What does it mean to say a DOG is unbalanced? What is the significance of this?
If it is unbalanced, convolution will yield a nonzero result. This is basically meaningless. It means: there’s something edge-like somewhere near me (within the scope of the particular DOG at that particular place)
What does it mean to say a DOG is balanced? What is the significance of this?
When balanced, the convolution algorithm yields a result of 0. This means, to a first approximation, there is no edge within the scope of the DOG.
How can we use DOGs to detect zero crossings in a luminance map?
Look at DOG responses at different locations over (across) the image
In this way, DOGs compute the second derivative of a luminance map
What are some characteristics of the DOG response?
~Vary with location ~Insensitive to uniform luminance ~Insensitive to linear changes ~Sensitive to spatial scale -neurons downstream know spatial scale of edge depending on which DOG detected the edge
What is disparity? Why is it a useful cue to depth?
~Disparity: a single point in the world can project to different locations in the two eyes
-Location in right - location in left
This is cool because…disparity varies with depth (distance from viewer) so it is useful for computing differences
What is the horopter?
Fixation point resides on an imaginary sphere…
-Horopter: collection of points out in the world that are as far away as the fixation point
Why is there no disparity at the fixation point?
The point at fixation projects to the same location (the center) in both eyes
So, it projects to the fovea in both eyes.
Since it is the same point, there is no disparity between the two eyes
Let’s say we have a fixation point, F. Point A is beyond fixation. What kind of disparity will we experience with point A, and why?
~Uncrossed disparity: appears to the left of fixation in the left eye and to the right of fixation in the right eye
-this is uncrossed because the image projects to the retina upside-down and backwards –> uncrossed from the perspective of the viewer
If a point experiences uncrossed disparity, where is it in relation to the horopter? Is disparity positive or negative?
If uncrossed –> further away than the horopter
Disparity is negative
Let’s say we have a fixation point, F. Point B is closer than fixation. What kind of disparity will we experience with point B, and why? Is disparity positive or negative?
~Crossed disparity: appears to the right of fixation in the left eye, and to the left of fixation in the right eye
-projects to the left of center of the left eye and right of center of the right eye
Disparity is positive
Let’s say we have a fixation point, F. Point C is on the horopter, but to the left of fixation. What kind of disparity will we experience at point C, and why?
~The disparity at Point C is 0
-Point C would project to the right of center in both eyes
What is the correspondence problem the visual system faces?
There are almost always multiple features in both eyes…so, which feature in the left eye corresponds with which in the right eye?
What are the constraints on the correspondence problem?
~Compatibility: any feature (area of local contrast) in the left eye corresponds to a similar feature in the right
-e.g., red spot in LE corresponds to red spot in RE, not green spot in RE
~Uniqueness: each feature on the left goes with at most one feature on the right, and vice versa
~Continuity (i.e., smoothness): depth, and therefore disparity, tends to change smoothly everywhere
-except at boundaries between surfaces
What is the representation the visual system uses to solve the correspondence problem?
Matrix that represents the output of ganglion cells in a slice through each retina
- Indicates where features are present in both eyes
1) Retinal units: active when feature is present, inactive when no feature is present
2) Correspondence units: active when left eye location (the columns) corresponds to right eye location (rows)
3) Diagonals correspond to specific disparities - Crossed disparity goes from furthest right column to furthest left
- Uncrossed disparity goes from bottom row to top
What is the algorithm used by the visual system to solve the correspondence problem? Generally, how are the constraints implemented?
~Constraints implemented by allowing correspondence units to interact: excite/inhibit one another
Algorithm:
1) Compatibility: excite any correspondence node that sits at the intersection of two active retinal units (i.e., features)
-Conjunction of retinal units is like an “and gate”
2) Uniqueness: correspondence nodes in the same row or column inhibit one another (silence inconsistent match-ups)
3) Continuity (smoothness): adjacent correspondence nodes in the same diagonal excite one another
~Try to satisfy all of these constraints using parallel constraint satisfaction
What is parallel constraint satisfaction? Where are the excitatory connections?
~Exchange excitation and inhibition in parallel until nodes reach stable activations
-in layman’s terms…initially, there are a bunch of units that represent the hypothesis. some are consistent, some are inconsistent. let them talk to each other. over time, the ones most consistent with evidence are active at the end. the constraints are maximally satisfied, and the representation settles in to this state.
~Excitatory:
-from retinal units to correspondence units
-between correspondence units (on diagonal)
How is the solution to the correspondence problem implemented in the hardware of the brain?
~This happens in V1
~Some neurons in striate cortex (V1 and V2) are selectively responsive to specific disparities
-zero, crossed (near), and uncrossed (far) disparities
What is inverse optics? (related) What is the goal of the visual system?
~Optics = light, image formation, turning 3D input into 2D representations
~Inverse optics “undoes” the optical process
-recover 3D world that gave rise to that 2D image
^this is the problem the visual system is trying to solve
Why is inverse optics a difficult problem for the visual system to solve?
It’s fukkin unsolvable! The info in the 2D image underconstrains the 3D interpretation
^you necessarily lose some information when squashing the 3D world into a 2D image
How can the visual system work around the unsolvability of the inverse optics problem?
~Visual system makes intelligent guesses using heuristics:
although any image is in principle consistent with an infinite number of worlds, some of these interpretations are more likely to be correct
~Visual system assumes world most likely to create the image
What is the non-accidentalness assumption?
Visual system assumes that every image’s features come from a world that is likely to produce it rather than it being an accident of viewpoint
Give an example of how relative size could be used as a monocular depth cue.
O o .
These circles could be at the same depth, and just different sizes. Or, they could be the same size and receding in distance.
Give an example of how linear perspective could be used as a monocular depth cue.
|\
Could be two nonparallel lines at the same depth, or parallel lines that appear closer together as the recede into the distance. Think roads far off.
Give an example of how texture gradient could be used as a monocular depth cue.
If the mean size of the texture gets smaller, that means the surface recedes in depth.
How is shading helpful as a monocular depth cue?
Only works if you know where the illumination comes from. But, once you do, illumination is assumed to be a parameter and applied to all objects in the image.
How is interposition useful as a monocular depth cue?
The object that occludes the other is perceived to be in front. It can also be a clue to where light is coming from (shadows)
How is motion parallax useful as a monocular depth cue?
When you are moving, closer things move faster across your retina than things farther away (provided those things are stationary)
Give an example of how familiar size can be used as a monocular depth cue.
If a baby appears to be the same size as a house in our vision, we know the baby is closer to us because a baby is much smaller than a house.
How can aerial perspective be a useful monocular depth cue?
Things closer to us are clearer, and things farther away (on the scale of kilometers) are hazier.
Human perception and cognition are characterized by sharp and ubiquitous capacity limits. Wouldn’t it be great if we could process everything at once?
Not really. Processing capacity limits are a side effect of the fact that we’re solving other problems nearly optimally.
One problem the visual system faces is how to allocate brain power to retinal input to maximize relevant information processing. How do we solve it?
~Devote hella power to central foveal vision (lots of cones –> super high acuity) and move eyes around
-maximizes cortical processing in whatever stimuli is of current interest
What is meant by the term “cortical magnification”? What are pros and cons?
~The fovea is way over-represented in the cortex – it gets more disproportionately more brain than other parts of the retina
Pro: lots of computing power for fovea
Con: need to move eyes a lot
-but, eye movements are cheap
–> in its hardware, the brain is selective about what gets processed!
What is meant by the term “retinotopic mapping”? Describe how this is related to cortical magnification. (Remember dat munkey study)
~Retinotopic mapping: neurons in adjacent portions of V1 represent adjacent parts of the real world
-The fovea can only fixate on a small portion of a visual scene, and that portion’s representation is “blown up” out of proportion in V1
Why is selecting information important for the visual system?
~Any given task requires only a subset of the available information, and different tasks require different subsets
~Furthermore, different tasks require different combinations of information, potentially from different sources
-Must attend to some parts of stimulus and ignore others
~Information is necessarily confounded in the world – every stimulus contains every kind of information, but you only need to attend to some of it
~Most tasks require you to integrate those pieces of information that are relevant
Describe the selection problem that the visual system faces. What makes it possible to solve?
- Different tasks require different combinations of information
- Information is confounded in the world
–>So, must select relevant and filter out irrelevant info
Solvable because: independent pieces of information are represented independently
-you can perceive some things separately, but not all
Describe the binding problem with respect to attention.
- Most tasks require you to integrate those pieces of information that are relevant
- If there are multiple objects, then you need to keep track of how properties are bundled together into objects
What are the four fundamental functions of attention?
- Selection: which properties, of which objects
- Binding: which properties go together to form objects, which properties to integrate for the task at hand
- requires a binding tag of some sort
- Mapping: for what porpoise – i.e., where does the information go?
- Enhancement: “better” processing (e.g., more precise/refined representation) of attended information
What is the implication of a finite number of binding tags?
We have a processing capacity limit!
Describe the results of Treisman and Gelade’s parallel search condition.
(when all distractors differ from target on one dimension)
~Search slopes near zero –> little or no cost for additional distractors
~Target absent slope = target present slope
SO…feature detection is automatic/parallel all over the visual field
Describe the results of Treisman and Gelade’s conjunction search condition.
(when some distractors differ from target on one dimension but have another in common, and another set of distractors differ conversely)
~Search slopes nonzero –> cost for additional distractors
~Target absent slope = 2x target present slope
–> self-terminating search
SO…feature binding is controlled, serial, and requires attention
Describe the mapping problem of attention.
~Another manifestation of the selection problem
~Not every part of the brain is directly connected to every other part –> means that not all computations happen automatically (so we can decide which response to make)
~Attention as a “switchboard” for passing information from one place to the next - decides where output of various processes go
~Relation to automatic vs. controlled processes
Describe characteristics of automatic processes.
~Require little attention ~Take less capacity (e.g., WM) ~Ballistic (once started, hard to stop) ~Run in parallel ~Interfere with less automatic processes
Describe characteristics of controlled processes.
~Require attention ~Take more capacity ~Not ballistic ~Serial with respect to one another ~Interfere little with more automatic processes
What are automatized processes?
Processes you’ve made automatic by practicing them a lot
Why do we need a flexible basis for mapping stimuli to response? What is this basis?
~There are some stimuli you respond to in the same way every time (e.g., word reading, object recognition)
~For the rest, you need to be able to map stimuli to responses flexibly
The basis is attention!
Describe the tasks in the Smith and Magee (1980) study.
Stimuli were a picture of some object with a word superimposed over it. The word was either consistent or inconsistent with the picture.
In two conditions, the word was the relevant part of the stimulus:
1) read word, ignore picture – requires you to access name
2) categorize word – requires you to access meaning
In the other two, the picture was the relevant part:
3) name picture, ignore word – access name
4) categorize picture – access meaning
Describe the results of the Smith and Magee (1980) study.
~If the task is naming a picture, then performance is fucked up by the presence of an inconsistent word
–>Word naming is more automatic than picture naming – words activate names automatically, and pictures do not
~If the task is classifying words, then performance is fucked up by the presence of an inconsistent picture
–>Picture classification is more automatic than word classification – pictures activate meanings automatically, words do not
Describe the results of the Sternberg (1966) study.
~Slope is nonzero –> serial search
~Target present slope = target absent slope –> non-self-terminating search
Describe the setup of the first Schneider and Schiffrin (1977) study.
~Manipulated whether the mapping from probe to response was consistent or variable
- Consistent mapping: any given probe always got the same response –> makes it possible to learn a direct line between stimulus and response
- response is either always a target or always a distractor (not necessarily on every trial, but on every trial it is present)
- Variable mapping: any given probe sometimes gets one response, sometimes gets the other –> impossible to learn direct probe-response mapping
What are the main variables examined in the first Schneider and Schiffrin (1977) study?
~Memory set size
~Visual search display
~Consistent vs. variable response mapping
Describe the results (after practice) of the first Schneider and Schiffrin (1977) study.
~Varied condition: slopes increase with size of visual set
^need to do more comparisons
~Consistent condition: slopes flat in frame size and memory set
–> visual search and memory search became automatic!
Descrive the additional evidence of automatic visual search in the second Schiffrin and Schneider (1977) study.
~First, train people with consistent mappings until they become automatic
~Then, use consistently-mapped targets as distractors in later displays
~Results: distractor pops out, interfering with detection of target: subjects’ ability to detect target in the presence of a previously consistently mapped target decreases by 22%
Putting all the evidence together, how do we make controlled psychological processes automatic?
Make the same stimulus-response mapping consistently!
Describe the characteristics of the ventral pathway.
~In charge of determining WHAT is out in the world
-for categorization, inference, deciding if you want to interact with something
~Generally, ignore what you need to know for dorsal pathway activities
Describe the characteristics of the dorsal pathway.
~In charge of determining WHERE things are in the world and HOW to interact with them
~Damage to this pathway causes difficulties with attention
~Generally, ignore what you need to know for ventral pathway activities
Describe the most common symptom of unilateral neglect. What does it result from?
~Characterized by an inability (or relative inability) to attend to one side of space
- results from parietal damage to contralateral side of brain
- most common symptom is failure to notice stimuli contralateral to damage
Give some evidence that unilateral neglect is more than just a failure to notice stimuli contralateral to damage.
~It’s object-centered – neglect rotates with objects
~Applies to memory
-when asked to recall the buildings of a plaza, patients will recollect all the buildings on one side – the side without the damage
What (attentionally) does unilateral neglect result from?
~Failure to disengage attention from the currently attended region of (space or object-centered) space
-when asked to cross out all lines on page, patients don’t cross out lines on side with damage
vs-when asked to erase all lines on page, patients erase all the lines because there’s no stimuli left to disengage attention from
What does unilateral neglect show us about the relationship between the visual system and attention?
You can’t have an intact visual system without an intact attentional system.
What does Balint’s syndrome usually result from?
Bilateral damage (dorsal pathway damage)
What are the symptoms of Balint’s syndrome?
~Simultaneous agnosia: can only see one object at a time
-usually can switch attention with motion (because motion trumps other stuff)
~Ocular apraxia: can’t voluntarily guide eye movements to a new location of visual fixation
~Spatial disorientation: inability to locate objects in space (relative to one’s self and one another)
~Optic ataxia: inability to reach to objects in space
Is Balint’s syndrome object-centered or space-centered?
Object-centered!!
