1. Fruit Fly Learning: Operant and classical conditioning
Björn Brembs, Berlin
Fruit Fly Learning: Operant and classical conditioning

2. Classical conditioning

3. Operant conditioning
Skinner-Box

4. Operant conditioning
Model
Critical research must address both stages!

5. Stage 1: The Actor

6. The Drosophila Flight Simulator
Stage 1: 1/11
The Drosophila Flight Simulator

7. Stage 1: 2/11
Fly Turning Behavior

8. Fly Turning Behavior Stage 1: 3/11 Open-loop Model (Robot-Hypothesis)
Alternative Hypothesis:
Question: Is spontaneous behavior under neural control?

9. State Space Reconstruction
Stage 1: 4/11
State Space Reconstruction
Coordinate Embeddings
Time series: 80, 60, 40,……
Embedding Dimension: 3
Three data points determine the coordinates of a 3D vector:

10. Is Turning Behavior Random?
Stage 1: 5/11
Is Turning Behavior Random?
Analyzing inter-spike-intervals
Geometric Random Inner Products: GRIP
If Drosophila turning behavior is not random, how is it distributed?
Lévy distributed!
All calculations: Alexander Maye, UKE Hamburg

11. Order in Turning Behavior
Stage 1: 6/11
Order in Turning Behavior
Analyzing inter-spike-intervals
Correlation dimension:
What is the probability to get the original CD
with shuffled data?
All calculations: Alexander Maye, UKE Hamburg

12. Nonlinear Forecasting: S-Maps
Stage 1: 7/11
Analyzing inter-spike-intervals
Use one part of the series to predict another one
Plot the correlation between the two parts
Use a weighting parameter to describe the increasing nonlinearity of the models used for the prediction
If the correlation increases with the weighting factor, the output is nonlinear
All calculations: Alexander Maye, UKE Hamburg

13. Nonlinear Forecasting: S-Maps
Stage 1: 8/11
Analyzing raw yaw torque data series
Logistic map:
Couplings:
All calculations: Alexander Maye, UKE Hamburg

14. Stage 1: 9/11
Logistic Map

15. Stage 1: 10/11
Logistic Map

16. Stage 2: The Critic

17. Switch Mode Learning Stage 2: 1/2
Two predictors precede the heat (US): color (CS) and behavior
Important: each predictor can also be learned separately

18. Hierarchical Interactions
Stage 2: 2/2
Hierarchical Interactions
Analyzing sw- and yt-learning
How do learning mutants perform in these paradigms?
Exchanged contingencies
sw-training; yt-test
Yoked colors

19. Take Home Message Operant conditioning consists of two stages.
Stage 1 actively initiates a variable range of behaviors.
Stage 2 compares outgoing behaviors with incoming sensory data (output/input transformations).
Operant conditioning differs from classical conditioning on the behavioral, neural and molecular level.
Operant and classical conditioning interact hierarchically in “composite conditioning” to accomplish maximum learning efficiency in natural situations.