Predictive processing Flashcards
Prediction error minimization
- Build model of the world, H, based on sensory input
- Evaluate H by taking action, thereby getting a new O, using that to update the probability of H
- Keep taking action to minimise the error on H to build confidence in H
- If new O indicates that H’ is more probable, abandon H, and evaluate H’ by taking action
Bayes rule/inference
P(H│O)=(P(O│H)·P(H))/(P(O))
Sensory input
What we perceive (bottom up)
Expectation
What we expect to perceive
Bottom up processing
An outer stimulus (that often was uninspected) that goes from the senses to the brain
Top down processing
The brain expects something
Hierarchical processing
There are different layers which become more and more abstract and when we perceive something it starts with the least abstract one (our senses)
Likelihood
P(O|H)
The probability of observing something given the hypothesis
Binocular rivalry
We can only perceive one thing at a time, so when we see two overlapping images with a face and a house we can only perceive one of them at a time (you can only focus on one thing and the other things will be “blinded out”)
Rivalry is characterized by this very dramatic change in actual visual consciousness
Today’s posterior is tomorrows prior
We always update our priors and beliefs
Solution to symbol grounding problem
No longer symbols
Feedforward error signal
Bottom up processing. When we perceive something unexpected
Main assumption of predictive processing
“a […] substantial view based on the rather uncontroversial idea that the brain is involved in information processing, and that information theory is cast in terms of the probability theory from which Bayes’ rule is derived”
The brain is only concerned with minimizing prediction error
Normative notion
How it should be and not how it is
Díaz-Caneja stimuli
Reveal that we update our prior probability/hypothesis.
In 1928 Emilio Diaz-Caneja (Diaz-Caneja 1928) discovered that if the two images are cut in half and combined such that one eye sees, for example, half a house and half a face, and the other eye sees the other halves of the house and the face, then there is not rivalry between what is presented to each eye, there is instead rivalry between the full, uncut images of the face and the house
Active inference
We sample sensory input in the light of the model we hold about the world. We should expect that the brain minimizes prediction error by changing its position in the world and by changing the states of the world, both of which will change its sensory input
Action in this sense of testing perceptual models is therefore a moment of prediction error minimization
Perceptual inference
Is done following Bayes’ rule to infer causes of sensory input. Minimization of the variation of the expected prediction error. A matter of selecting and adjusting hypotheses about the world in response to the prediction error they engender
External interpretation
Interpretations that come from experience
Phenomenology
An approach that concentrates on the study of consciousness and the objects of direct experience.
Neuroanthropomorphism
Inappropriately imputing human-like properties to the brain and thereby confusing personal level explanations with subpersonal level explanations
Functionalist
Mental states are defined by a functional role that specifies a certain kind of input-output profile, given a certain internal state
Causal inference
The process of determining the independent, actual effect of a particular phenomenon that is a component of a larger system
Inferential
Characterized by or involving conclusions reached on the basis of evidence and reasoning.
Invariant representations
Representations that do not change
The perceptual hierarchy
The basic idea for the hierarchy of perceptual inference is that for every level of the hierarchy we probe deeper into the causal structure of the world
Empirical Bayes
The prior expectations are pulled down from what has previously been learned best at higher levels of the hierarchy
Attention allocation
Updating working memory, and language processing are interdependent cognitive tasks related to the focused direction of limited resources, refreshing and substituting information in the current focus of attention, and receiving/sending verbal communication, respectively
Agency
Important, without it we would be stuck at our starting point
To improve our position in the world and that we, of course, use the way we perceive the world to inform and guide agency
Perceptual inference
A matter of selecting and adjusting hypotheses about the world in response to the prediction error they engender
James-Lange theory
Feeling states arise as responses to states of bodily arousal including rise in heart rate, perspiration, and so on
E.g. “I cry therefore I’m sad”
Interoceptive
The feeling of knowing what is happening in your body
Introspection
We attend to our own mental states
Introspective dissonance
Introspection seems both accessible and certain, and inaccessible and uncertain.
Emotion
Perceptual inference on our internal states
Privacy of self
Makes my estimate of the state of affairs independent of the estimates of others, meaning that we can use the estimates together socially to make even better estimates
The self
A self-model is needed to predict how acting on the world results in new sensory input
Best inference
The inference that minimizes prediction error
Marginal probability
P(O): The sum of the probabilities of O conditional in all hypotheses
