Scribd
Upload
21 views

Predictive Coding

Uploaded by

acol0074
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
21 views

Predictive Coding

Uploaded by

acol0074
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Overview

Thursday, 27 April 2023 1:15 PM

- The brain is constantly gathering and updating a mental model of our environment
○ It uses information available to generate predictive models of what will happen next
- The brain is optimised to predict surprises
○ Surprises can be changes, violations of assumptions or predictions, irregularities and unexpected events
- The goal is to minimise surprise by maximising the accuracy of our predictions
○ Thus we continually update our mental model
○ It is more efficient to respond to only the changing/surprising stimuli in the environment rather than all elements
Mismatched negativity
- A waveform reliably present on an EEG when expectations about stimuli are violated
○ Present in superior temporal and inferior frontal gyrus
- Demonstrated by the oddball paradigm
○ Usually done with audio stimuli
○ Plays a participant a standard tone over and over and then a deviant tone
▪ The EEG waveforms are recorded

- The waveform of MMN (mismatched negativity) is always the same


○ It has a characteristic dip immediately after the violation
○ Occurs between 150-250 ms after the violation
▪ Time of onset depends on the modality of the stimuli and the rule complexity

- MMN can be triggered by violations of expectations in stimuli of many modalities


○ Visual, audio, somatosensory
- Changes in physical attributes of stimuli that cause MMN responses include
○ Pitch
○ Duration

○ patterns

Predictive coding Page 1


▪

○ Unexpected sequence changes

▪ The expected uncertainty would not generate an MMN because from the get go, the participant knows it's going
to be different, which is not the case in the unexpected uncertainty group
○ Location
▪ Both visual and auditory
- Deviations from normal distributions of audio tones have been demonstrated to produce MMN
○ In both narrow and broader distributions

○ Suggests that it may be a statistical cognition?


Why does MMN occur
- Local adaptation hypothesis
○ The brain areas adapt to the repeated stimuli, therefore violations of expectations cause large effects
- Model adjustment hypothesis
○ The MMN occurs as the brain is comparing the new stimuli to the memories of the older, more expected stimuli
○ comparison between sensory input and a memory trace of previous input

Predictive coding Page 2


Clinical implications
Monday, 1 May 2023 1:30 PM

Schizophrenia
- MMN in reduced in schizophrenia
○ It is the most robust biomarker of schizophrenia

- MMN reductions cause issues retaining auditory stimuli in working memory


○ Prevents pattern recognition in language or auditory stimuli
- MMN could be used as a screening tool
○ MMN reductions predict 12 month onset of psychosis in high risk populations

- MMN reductions in different brain areas can predict symptom severity of hallucinations, anhedonia, avolition and
attentional difficulties
○ Brighter = more contribution to that symptom

Predictive coding Page 3


○

- The continuum, dimensional approach to psychosis and psychotic symptoms in the general population is supported by
MMN data

- Psychosis patients have less connectivity between brain regions

- White matter connectivity in the audio pathways between the left and right halves of the brain is associated with psychosis
experiences in non-clinical population
○ Loss of integrity in what matter = more psychotic symptoms
Coma
- MMN is reduced in patients in coma
○ The presence of the MMN can predict recovery and waking
○ 9/10 (90%) comatose patients with MMN eventually awakened
- MMN responses fade with consciousness

Predictive coding Page 4


Bayesian Inference
Monday, 1 May 2023 7:30 PM

- Making meaning out of sensations in the face of conflicting, ambiguous or missing information
○ Perception is not passive, it is an active process of making sense of our sensory information into something we
understand
- Perception is influenced by our previous experiences
○ Memory acts as a template
- According to the Bayesian Brain hypothesis, sensory information is represented probabilistically
○ We use our prior knowledge to resolve uncertainly and guide our decisions
Prior knowledge and likelihood decisions
- Prior knowledge and the known likelihood of events occurring influences what we think will eventually happen
○ Prior knowledge is what has happened in the past
○ Likelihood is the current probability of the thing happening
- Knowledge but not likelihood
○ When our prior knowledge, experience or learning is very strong, then the predicted outcome in an ambiguous
situation will more strongly reflect the influences of previous learning rather than the likelihood of that event occurring
- Likelihood but no knowledge
○ When we have strong inclinations about the likelihood of an event and our prior knowledge is weak, the predicted
outcomes will be much closer to the likelihood
- When likelihood and prior knowledge are evenly represented, decisions will situate themselves between the two

Brain areas
- Greater likelihood uncertainty response in the superior occipital cortex
- Greater prior knowledge uncertainty response in the putamen, amygdala, insula, and orbitofrontal cortex
Processing Hierarchies and predictive coding
- Bayesian inference is supported by processing hierarchies
○ Prior knowledge gets communicated to lower level processing centres for analysis of sensation and to create
perception
○ Predictive coding proposes that Bayesian inference is implemented in the brain via top-down prior knowledge being
compared with sensory information from downstream
- Processing hierarchies support predictive coding
○ Mismatched negativity responses are sent upwards to update prior knowledge for Bayesian inference
Autistic model
- Autistic individuals have similar prior knowledge to neurotypicals. But they rely more on and have a more accurate use of
sensory input. So their perceptions are based more on input information than previous knowledge

Predictive coding Page 5


Predictive coding Page 6

You might also like