Title: How facilitation influences an attractor model of decision making
1How facilitation influences an attractor model of
decision making
2Outline
- Facilitation and depression in cortical neurons
- How does it work and what is it good for?
- 2. Facilitation and working memory
- 3. Facilitation and decision making
- Sequential decision making
- Parallel decision making
- 4. Outlook
Lunch
3Neurophysiological mechanism
Idea - docked pool of vesicles containing
neurotransmitter - each released vesicle is
replaced with a time constant td. -
intracellular calcium responsible for release
probability of vesicles
Synaptic depression The neuron fires more
rapidly than vesicles can be replaced. ? Vesicle
depletion
- Facilitation
- APs arriving in the presynaptic terminal cause
the accumulation of calcium ions. - ? increase of released vesicles/AP
From Wang Y et al. 2006
4Possible benefits of facilitation and depression
- Synaptic Filter (low-, high- or band-pass)
- Decorrelation
- Burst detection
- Stabilize against noise (Gain Control)
- Persistent activity
- Depression
- Persistent activity can be maintained with
relatively low firing rates (e.g. 15 Hz). - Facilitation
- May increase the strength of connections above a
critical threshold necessary for the stability of
persistent firing. -
? Brief periods of synaptic activity may
transiently shift connected cortical neurons into
a state in which recurrent excitation is
sufficiently strong to support persistent
activity.
5Mostly depression in the cortex?
Short-term synaptic depression between pyramidal
neurons has been observed in all cortical areas
that have been examined (mainly V1, sensory
areas).
- Special properties of PFC
- Heterogeneity of synaptic dynamics (40-50
facilitation dominant). - Facilitation is often masked by high initial
probability of release. - Augmentation with time constants of several
seconds.
From Wang Y et al. 2006
6Why in PFC? Working memory!
7Why in PFC? Working memory!
- Synaptic Theory of Working Memory
- Mongillo et al. (2008) Science 319, 1543
- Memory is maintained by short-term synaptic
facilitation. - (mediated by increased residual calcium levels
at the presynaptic terminals)
- Object Working memory dominated by recurrent
inhibition - Brunel and Wang (2001) J Comp Neurosci 11, 63-85
- Working memory based on persistent activity
maintained through recurrent NMDA synapses with
long time constants. - (Synapses are permanently enhanced by previous
long-term potentiation)
x available resources (vesicles) u fraction of
resources used by each spike (residual calcium
level) tF 1.5 s gtgt tD 0.2 s
8F-D spiking network
- integrate-and-fire neurons
- (inhibitory and excitatory)
- structured network
- (p selective populations each encoding one
memory item) - E-E connections weights are multiplied by
u(t)x(t) (0 gt ux gt 1)
From Mongillo et al. 2008
? Different regimes of working memory depending
on background input
9F-D spiking network
- Memory maintained synaptically
Memory maintained by persistent firing
From Mongillo et al. 2008
10Temporary vs. permanent synaptic enhancement
- Advantages of strengthening synapses
temporarily - metabolically more efficient
- (conserve synaptic resources during periods of
baseline activity) - greater flexibility and control
- (synaptic facilitation is activity dependent and
can vary in a graded fashion) - regulation of persistent activity by altering
presynaptic release probability through
neuromodulators (e.g. dopamine) - Quiescent memory ? decoupling from other brain
areas
11So many advantages why not try in decision
making?
12Facilitation and sequential decision making
- Task Comparison of two vibrotactile stimuli
applied sequentially
Problem Model of Wang 2002 (attractor network of
two competing populations) does not work in
sequential decision making
Deco et al. in preparation
Experimental findings partially differential
neurons in VPC respond to f1, show no persistent
activity, but ramp up at the end of the delay
period and are influenced by f2 during the
decision period
13Facilitation and sequential decision making
Specific input
f2
f1
delay
Global attention signal
Deco et al. in preparation
14Facilitation and sequential decision making
- During application of f2, the level of activity
will depend on the synaptic history (uf1) and on
f2.
Together with the neurons that encode just f2, a
decision of f1 lt f2 or f1 gt f2 can be reached
with a standard attractor model of decision
making.
f1 gt f2
f1 lt f2
Deco et al. in preparation
15Facilitation and parallel decision making
- Task Gather evidence (dot motion) and decide
between multiple targets
Albantakis Deco 2009
- Facilitation
- Network performance and decision dynamics are
mainly preserved. - But greater flexibility in weights!
16Outlook
Idea Use facilitation (or combination of
facilitation and depression) for experimental
findings that are currently not completely
accounted for by the multiple choice model
without facilitation
- Log Odds (coherence-dependent difference between
in and out choice) - Extend the multiple choice model to more than 4
choices -
17Thanks
to my supervisor Prof.Gustavo Deco Consolider
and to everybody for listening!
18Facilitation and parallel decision making