Title: Opponent interactions between serotonin and dopamine
1Opponent interactions between serotonin and
dopamine
We will talk
- About Dopamine and Serotonin
- About TD learning
- How these neurotransmitters interact in TD
learning
- Review of D.Daw, S.Kakade and P. Dayan
paperDavidov Dmitry
2Dopamine
One of most researched neurotransmitters
- As shown
- Strong evidence for being related to reward in TD
learning system
3Serotonin (5HT)
- Most mysterious of the main vertebrate
neuromodulators
4Dopamine Serotonin controversyEvidences
- Pharmacological
- (Dopamine involved in behaviors that serotonin
inhibits) - Microdialisis
- Electrophysiological
5Appetitive Vs Aversive system
- A lot of evidences for existence of two separate
systems
Aversive
Appetitive
Stopping delivery of a long sequence of rewards
can also be aversive
As well as stopping delivery of punishments can
be appetitive
X
X
6How to predict future?... - TD learning
modelFor each trial
- Functions
- V(t) Value
- r(t) reward
What is needed is to approximate V(t) - Actually
we need to update V(t) in light of reward r(t)
Phasic component of error signal
7Problematic point
- How to represent negative errors?
- Pauses in neuronal firing?
- Baseline is extremely weak
- No evidence that downstream areas can detect
such pauses
8Another point
- Outside of laboratory trials are not episodic,
but all life is like one continuous trial
Solution
Use DIFFERENTIAL values V(t) Summed
difference between observed reward and some
average reward r
9How it should look like?
X
Reward
Extinction
?Net Responce?
Phasic Dopamin
Tonic Serotonin
10Symmetry wins?
- Strange fact
- Elevated dopamine in response to aversive
stimuli
Suggestion
Dopamine and serotonin are oposite in both
actions!
11Work to DO and OVER-simplifications
- How do the serotonin and dopamine systems
- actually divide their function?
- Are there any significant serotonin phasic
responses? - Is learning rate actually fixed?
- Adaptive learning rate?
- Are there only 2 time scales?
- Hierarchy of time scales?
12Conclusions
V
V
- Dopamine and Serotonin activity are likely to be
involved in a kind of TD learning system - Phasic and Tonic signals play distinct roles
- The functions of Dopamine and Serotonin may
mirror each other