The Leaky Integrator Model

1
The Leaky Integrator Model

• for lever-pressing mice
• Dan and Jinsook

2
The Experiment

• Mouse wants happy center stimulations
• Two levers
• Different probabilistic rewards
• Fatigue makes switching strategy important
• Reward probabilities change in middle of trial
  (which lever has higher reward switches)

3
Why is this interesting

• We want to understand how strategic decision
  making goes down in the brain
• The leaky integrator model worked well with
  monkeys making saccades
• Here rats do what they want when they want, so
  its more complicated and realistic How does the
  model hold up?

4
The Leaky Integrator Model

• Internal expectation function
• Add to function when press -gt reward, subtract
  when press -gt no reward
• Event memory degrades exponentially
• Resets whenever mouse switches
• We choose memory half-life of 1 sec, initial
  expectation 1, reward 1, noreward -1

5
Mice are bad at calculus

• Switch much more often when expectation function
  is low, but often dont switch when its low

6
Better than matching?

• Next step is to see if it improves modeling
  accuracy over matching