Transcript summary - Marieke Rohde

The Function of Synchrony
Marieke Rohde
Reading Group DyStURB
(Dynamical Structures to Understand Real Brains)
Structure
1. Sound recognition by transient
synchrony. (Hopfield & Brody)
2. Long distance synchronisation in Human
subjects (Rodriguez et. Al.)
3. Discuss!
1.) Hopfield, Brody: What is a
moment? (Puzzle and Answer)
Mus Silicium
• Short time integration in an artificial organism
Biologically plausible model, spiking neurons.
• Auditory task: one syllable recognition – short
time integration required to "represent" the
world.
• Mastered robustly
Mus Silicium - Anatomy
• Layer 4:
– 50% inhibitory, 50% excitatory
– Lots of cells and connections
• no delays, no plasticity
• Sensors: cells are frequency
tuned and respond to
– Onsets
– Offsets
– Peaks
• Transient decay of neural
activity at different decay
rates.
Mus Silicium - Anatomy
The alpha and beta neurons from „cortical layer 4“ exhibit
the same properties as the sensory neurons!
Mus Silicium: Responses
• Gamma cells: highly
specific to learned
syllable.
Mus Silicium: The Solution
General Principle: Transient synchrony of APs to „signal“ recognition
• Representation of time of a stimulus by different decay rates
• spatiotemporal patterns: Convergence of firing rate of decaying currents.
• Same rate neurons (coupled oscillators) tend to synchronise. (set weights
accordingly)
• Detection by cell with small time constant
• Invariant to time-warping (rescaling in time), delays and salience
Mus Silicium: The Solution
• 800 lines (different stimuli and
decay rates) from area A
project on an excitatory and an
inhibitory cell
• Training = find set of coinciding
neurons on pattern and
mutually couple them
(excitatory and inhibitory)
• Balance between excitation
and inhibition, to assure
network input current from
outside.
• Connect whole set to a gamma
neuron, to yield a reaction.
Mus Silicium
• Extensions:
– reactivation of sensors? (several, probablistic activation)
– Negative evidence. Destroy synchrony/detection.
•  Robustness against noise
– Multiple patterns: Phase transition n infty to general
synchrony
• Structure, not weights. Several structures conceivable
• Biological plausibility.
• Conclusion:
– A „Many are now equal“ operator.
– Model spiking networks if you want to explain the brain!
• How could you have guessed it?
2.) Rodriguez et.al.:
Perception's shadow: longdistance
synchronization of human brain
activity
Long Distance Synchrony
• 30-80 Hz oscillations
(gamma) synchronise
during a cognitive act.
(EEG MEG
measurements)
• Task: Recognition of a
degraded stimulus
(Mooney face)
Long Distance Synchrony: Methods
1. Detect induced gamma
response: "wigner ville time
frequency transforms“ of
single trials and average.
• first peak is known (much
stronger in perception
condition)
• second new, practically the
same for both conditions.
1. Phase synchrony:
– the phase synchrony profile is very different
from the gamma activity profile
– baseline: shuffled data.
– no perception remains close to baseline.
– perception: synchronisation, desynchronisation,
synchronisation (zero centered distribution of
phase lags).
Long Distance Synchrony:
Conclusions
• biological significance for cogntion
confirmed. (refutation to different
criticisms)
• High level, rather than low local feature
binding
• New finding: desynchronisation to prepare
for next synchronisation (destroy old
pattern).
• gamma activity != synchrony.
Discussion
•
Differences:
– Local vs. Global (+ role of delays)
– Detectors vs. Unknown function.
– Low level vs. High level
•
What methods to detect it in organisms?
– Phase lag: 0 or different?
– Time spans vs. every spike.
•
•
Synchrony - Asynchrony
What function could synchrony have?
– Attractive state (type of population code)
– Internal clock