how much about our interaction with – and experience of – our world can be deduced from

31
How much about our interaction with – and experience of – our world can be deduced from basic principles? This talk reviews recent attempts to understand the self-organised behaviour of embodied agents – like ourselves – as satisfying basic imperatives for sustained exchanges with our world. In brief, one simple driving force appears to explain nearly every aspect of our behaviour and experience. This driving force is the minimisation of surprise or prediction error. In the context of perception, this corresponds to (Bayes-optimal) predictive coding that suppresses exteroceptive prediction errors. In the context of action, simple reflexes can be seen as suppressing proprioceptive prediction errors. We will look at some of the phenomena that emerge from this formulation, such as hierarchical message passing in the brain and the perceptual inference that ensues. I hope to illustrate these points using simple simulations of auditory processing, with a special focus on repetition suppression in the context of the mismatch negativity and omission related responses. Predictive coding and repetition suppression Karl Friston, University College London

Upload: roman

Post on 04-Jan-2016

29 views

Category:

Documents


1 download

DESCRIPTION

Predictive coding and repetition suppression Karl Friston, University College London. How much about our interaction with – and experience of – our world can be deduced from basic principles? This talk reviews recent attempts to understand the self-organised behaviour of - PowerPoint PPT Presentation

TRANSCRIPT

Slide 1

How much about our interaction with and experience of our world can be deduced frombasic principles? This talk reviews recent attempts to understand the self-organised behaviour ofembodied agents like ourselves as satisfying basic imperatives for sustained exchanges withour world.

In brief, one simple driving force appears to explain nearly every aspect of our behaviour andexperience. This driving force is the minimisation of surprise or prediction error. In the context ofperception, this corresponds to (Bayes-optimal) predictive coding that suppresses exteroceptiveprediction errors. In the context of action, simple reflexes can be seen as suppressingproprioceptive prediction errors.

We will look at some of the phenomena that emerge from this formulation, such as hierarchicalmessage passing in the brain and the perceptual inference that ensues. I hope to illustrate thesepoints using simple simulations of auditory processing, with a special focus on repetitionsuppression in the context of the mismatch negativity and omission related responses.

Predictive coding and repetition suppression

Karl Friston, University College London

The anatomy of inferencepredictive codinggraphical modelscanonical microcircuits

Birdsongperceptual categorizationrepetition suppressionomission related responsessensory attenuationa birdsong duetOverview

Objects are always imagined as being present in the field of vision as would have to be there in order to produce the same impression on the nervous mechanism - von Helmholtz

Thomas BayesGeoffrey HintonRichard FeynmanThe Helmholtz machine and the Bayesian brainRichard Gregory

Hermann von Helmholtz Objects are always imagined as being present in the field of vision as would have to be there in order to produce the same impression on the nervous mechanism - von Helmholtz Richard Gregory

Hermann von Helmholtz sensory impressionsPlato: The Republic (514a-520a)

Bayesian filtering and predictive coding

changes in expectations are predicted changes and (prediction error) corrections

prediction error

Minimizing prediction error

Change sensationssensations predictionsPrediction errorChange predictionsActionPerception

A simple hierarchyGenerative models

whatwhereSensory fluctuations

Generative modelModel inversion (inference)A simple hierarchyDescendingpredictionsAscending prediction errorsFrom models to perception

Expectations:Predictions:Prediction errors:Predictive coding

Haeusler and Maass: Cereb. Cortex 2006;17:149-162Bastos et al: Neuron 2012; 76:695-711Canonical microcircuits for predictive coding

ThalamusArea XHigher vocal centreHypoglossal Nucleus

Prediction error (superficial pyramidal cells)Expectations (deep pyramidal cells)Perception

Action

David Mumford

Interim summary

Hierarchical predictive coding is a neurobiological plausible scheme that the brain might use for (approximate) Bayesian inference about the causes of sensations

Predictive coding requires the dual encoding of expectations and errors, with reciprocal (neuronal) message passing

Much of the known neuroanatomy and neurophysiology of cortical architectures is consistent with the requisite message passing

It is the theory of the sensations of hearing to which the theory of music has to look for the foundation of its structure." (Helmholtz, 1877 p.4)

Helmholtz, H. (1877). On the Sensations of Tone as a Physiological Basis for the Theory of Music", Fourth German edition,; translated, revised, corrected with notes and additional appendix by Alexander J. Ellis. Reprint: New York, Dover Publications Inc.,1954

Hermann von Helmholtz The anatomy of inferencepredictive codinggraphical modelscanonical microcircuits

Birdsongperceptual categorizationrepetition suppressionomission related responsessensory attenuationa birdsong duetOverview

Generating bird songs with attractorsSyrinxHigher vocal centertime (sec)FrequencySonogram0.511.5

Hidden causesHidden states

102030405060-505101520prediction and error102030405060-505101520hidden statesDescending predictionsAscending prediction error102030405060-10-505101520causal statesPredictive coding and message passingstimulus0.20.40.60.82000250030003500400045005000time (seconds)

Perceptual categorization

Frequency (Hz)Song a

time (seconds)Song b

Song c

Perceptual inference: suppressing error over peristimulus time

Perceptual learning: suppression over repetitionsSimulating ERPs to repeated chirps

100200300-10010LFP (micro-volts)prediction error00.20.4-505hidden states

Frequency (Hz)percept0.10.20.32000300040005000100200300-10010LFP (micro-volts)00.20.4-505

Frequency (Hz)0.10.20.320004000100200300-10010LFP (micro-volts)00.20.4-505

Frequency (Hz)0.10.20.320004000100200300-10010LFP (micro-volts)00.20.4-505

Frequency (Hz)0.10.20.320004000100200300-10010LFP (micro-volts)00.20.4-505

Frequency (Hz)0.10.20.320004000100200300-10010peristimulus time (ms)LFP (micro-volts)00.20.4-505

Time (sec)Frequency (Hz)0.10.20.320004000Time (sec)17Synthetic MMN

Last presentation(after learning)First presentation(before learning)1234500.511.522.53Synaptic efficacypresentationchanges in parameters123450123456Synaptic gainpresentationhyperparameters100200300-10010100200300-0.4-0.200.2100200300-10010100200300-0.4-0.200.2100200300-10010100200300-0.4-0.200.2100200300-10010100200300-0.4-0.200.2100200300-10010100200300-0.4-0.200.20100200300400-15-10-505101520primary level (N1/P1)peristimulus time (ms)Difference waveform0100200300400-0.6-0.4-0.200.20.4secondary level (MMN)peristimulus time (ms)Difference waveformSensory prediction errorExtrasensory prediction error

18Synthetic and real ERPs

A1A1STGsubcortical inputSTGIntrinsic connections12345020406080100120140160180200123450Extrinsic connections20406080100120140160180200presentationpresentation1234500.511.522.53presentationchanges in parameters123450123456presentationhyperparametersSynaptic efficacySynaptic gain

19The anatomy of inferencepredictive codinggraphical modelscanonical microcircuits

Birdsongperceptual categorizationrepetition suppressionomission related responsessensory attenuationa birdsong duetOverview

Sequences of sequences

Time (sec)Frequency (KHz)0.511.5

SyrinxHigher vocal centerSonogramArea X

omission and violation of predictionsStimulus but no perceptPercept but no stimulusFrequency (Hz)stimulus (sonogram)25003000350040004500Time (sec)Frequency (Hz)percept0.511.525003000350040004500500100015002000-100-50050100peristimulus time (ms)LFP (micro-volts)ERP (prediction error)without last syllableTime (sec)percept0.511.5500100015002000-100-50050100peristimulus time (ms)LFP (micro-volts)with omission

The anatomy of inferencepredictive codinggraphical modelscanonical microcircuits

Birdsongperceptual categorizationrepetition suppressionomission related responsessensory attenuationa birdsong duetOverview

ThalamusArea X

Higher vocal centreHypoglossal Nucleus

Active inference: creating your own sensationsMotor commands (proprioceptive predictions)Corollary discharge(exteroceptive predictions)

Active inference and sensory attenuation

Active inference and sensory attenuationMirror neuron system

time (sec)Frequency (Hz)percept1234567250030003500400045005000012345678-50050100time (seconds)First level expectations (hidden states)012345678-40-20020406080time (seconds)Second level expectations (hidden states)

time (sec)Frequency (Hz)percept1234567250030003500400045005000012345678-50050100time (seconds)First level expectations (hidden states)012345678-40-20020406080time (seconds)Second level expectations (hidden states)Active inference and communication

-20-100102030405060-20-100102030405060Synchronizationsecond level expectations (first bird)second level expectations (second bird)-20-100102030405060-30-20-1001020304050No synchronizationsecond level expectations (first bird)second level expectations (second bird)Mutual prediction and synchronization of chaossynchronization manifold

"There is nothing in the nature of music itself to determine the pitch of the tonic of any composition...In short, the pitch of the tonic must be chosen so as to bring the compass of the tones of the piece within the compass of the executants, vocal or instrumental. (Helmholtz, 1877 p. 310)

Helmholtz, H. (1877). On the Sensations of Tone as a Physiological Basis for the Theory of Music", Fourth German edition,; translated, revised, corrected with notes and additional appendix by Alexander J. Ellis. Reprint: New York, Dover Publications Inc.,1954

Hermann von Helmholtz Thank you

And thanks to collaborators:

Rick AdamsAndre BastosSven BestmannHarriet BrownJean DaunizeauMark EdwardsXiaosi GuLee HarrisonStefan KiebelJames KilnerJrmie MattoutRosalyn MoranWill PennyLisa Quattrocki Knight Klaas Stephan

And colleagues:

Andy ClarkPeter DayanJrn DiedrichsenPaul FletcherPascal FriesGeoffrey HintonJames HopkinsJakob HohwyHenry KennedyPaul VerschureFlorentin Wrgtter

And many others