mt 2 deck Flashcards
(97 cards)
bottom up procesing
piece together info coming infrom the world to make sense of it
predictive coding mocels
stress the idea that we dont have direct access to the world. we only get signals that are interpreted by our sensory systems
bayesian statisticians priors:
hypothesis about how likely things are in general and how lilely they are tp be true in the current situation
feedforward based mocel
signal or info gets apssed forward from one node to the next.
bottom up processing
feed forward (mostly) processes as info gets passed from V1 forward. neurons respond to features of objects in an increasingly large scale and higher levels of abstraction
feedforward models illustrate info that is
coming in from the world and going from basic sensory areas to high level integration areas of the cortex
treatise on psychological optics
brain is a prediction machine, perception is just unconscious inference (we infer the cause of a sensation in the world via its effects)
von helmholtz view of perception
generation of a best guess (inference) about the state of the world, in view of the data.
how do predictive coding models view the role of visual cortex neurons
predictions based on the probability the stimulus will have particular features. error detection- they respond to a mismatch between predicted signal and actual signal
generative model
model of the world that our brain produces whose business is to use what we know about how the world works to generate predictions (hypothesis) about what the object or scene in question is
higher level hypothesis
ex “there are no bears in pacific spirit park” or “there are bears in the mountains”
lower level hypothesis
influenced by higher level hypothesis. ex “if there is a bear there should be movement and sound”
lowest level hypothesis
specific to each modality (taste, touch, smell, hearing, vision). hypothesis are compared with incoming info from the senses. happens in the primary visual cortex (V1) /IOC. prediction sent down to all the specialized detectors of features
what happens at each level according to predictive coding?
there are representational units that encode expectations, or the probability of a given stimulus under the circumstances
what do representational units do?
send down their predictions of what they expect to receive from the lower level
what do error units do?
encode or read suprise (mismatch between predictions and bottom up evidence from the senses)
what happens to error signals
sent forward up to the next higher level where the expectations are adjusted
what happens if a mismatch between prediction and reality?
prediction error is generated
what happens to the prediction error
sent up the heirarchy, causing the revision of the hypotheses at the level above. if the next level up cant minimize the prediction error, then the prediction error gets pushed farther up the system
what does a higher level in the hypothesis levels mean?
more substantail revision of the hypothesis
when does preception happen
when the prediction error is minimized with the winning hypothesis forming the contents of the preception
basics of the egner paper
used an encoding approach to fMRI to empirically test predictions stemming from predictive coding models against feature-based models of object perception in the ventral stream
egner paper big picture question
does the predicitive coding explain visual objects recognition better than classic heirarchal feature-based model?
how was the big picture question in the egner paper studied?
by taking advantage of what we know about category selective populations of neurons in the fusifor face area (FFA) and parahippocampal place area (PPA)