A hierarchy of time-scales and the brain

1
A hierarchy of time-scales and the brain

• Stefan Kiebel

2
Overview
1 Biophysical models for EEG/MEG 2 Functional
model An agent just like the brain 3 Auditory
example
3
Overview
1 Biophysical models for EEG/MEG 2 Functional
model An agent just like the brain 3 Auditory
example
4
EEG and MEG Connectivity analysis
Biophysical model
Evoked response

Neural mass model
Network of nodes
Standard stimulus
Deviant stimulus
Exc IN
µV
PC
time (ms)
Inh IN
Sigmoid
Potential function

• David et al. (2006), NeuroImage Kiebel et al.
  (2006), NeuroImage, Kiebel et al. (2009), Human
  Brain Mapping

5
Biophysical modelling Applications
Auditory perception

EEG Evoked
EEG/MEG evidence of prediction error?
Prediction error of which predictions?
Garrido et al. PNAS (2007), J Neurophy(2009)
MEG Evoked
Schofield et al. PNAS (2009)
6
Functional role of network nodes
Biophysical model (MEG)

Model
Functional role of network nodes?
Brain data

Functional model
Amplitude (fT)
Input
Time (ms)
7
Overview
1 Biophysical models for EEG/MEG 2 Functional
model An agent just like the brain 3 Auditory
example
8
Meaning Hidden at slow time-scales
Single time-scale
Multiple time-scales


e1
e2
e3
e4
e1
e2
e3
e4
Fast
s1
s2
Slow

• Recognition
• non-robust
• no higher level representation

• Recognition
• robust by more constraints
• higher level representation

9
Speech example Fast and slow
frequency
time

• von Kriegstein et al. (2008), Curr Biol

10
Speech example Fast and slow
l
l
frequency
time
von Kriegstein et al. (2008), Curr Biol
11
Auditory recognition The brain challenge
Environment

Sound wave
Bayesian agent
Amplitude
Time (ms)
Online recognition/prediction
continuous dynamics
at multiple time-scales
at multiple time-scales
using continuous dynamics
expressing prediction error
12
Overview
1 Biophysical models for EEG/MEG 2 Functional
model An agent just like the brain 3 Auditory
example
13
Functional model of speech perception
Environment
Agent
Syllabic level
Phonemic level
Temporal hierarchy
Acoustic level
Online decoding
Sound wave

• Kiebel et al. (2009), PLoS Comp Biol, Friston.
  (2008), PLoS Comp Biol

14
Generative model Hierarchy of sequences
Syllables
Hidden states
2
Level 2
1
3
Phonemes
Hidden states
e
e
e
a
a
a
Level 1
i
i
i
o
o
o

• Kiebel et al. (2009), PLoS Comp Biol

15
Recognition of sequences
Environment
Agent
Syllables
Syllables
Phonemes
Phonemes
Sound wave
16
This means...
Hi
Hi
Bayesian agent
Hidden message at slow time-scale can be decoded
17
Deviations from phonotactic rules
Generative model of environment
Generative model of agent
2
2
1
1
3
3
e
e
e
e
e
e
a
a
a
a
a
a
i
i
i
i
o
o
o
o
o
o
18
Deviations from phonotactic rules
Recognized syllables
True syllables
Syllables
Phonemes
Syllables
Prediction errors
19
This means...
Hola
???
Bayesian agent
Prediction error deviations from expected
temporal structure
20
Predictions for experiments?

Environment

Sound wave
Input
Bayesian agent
Amplitude
Time (ms)
Online recognition/prediction
continuous dynamics
at multiple time-scales
using continuous dynamics
expressing prediction error
21
Conclusions
Auditory recognition/prediction can be modelled
by Bayesian online inference.
Input must be based on multi-scale temporal
hierarchy.
Outlook derive functional predictions for
experimental testing
22
Thank you
Karl Friston Jean Daunizeau Katharina von
Kriegstein