L15 - Linking Senses and Behaviour to Memory

Question 1

What are David Marrs 3 levels of analysis?

Answer 1

Computational problem
Algorithm
Physical implementation

Question 2

1st level - computational problem summary

Answer 2

What problem are you trying to solve?

Integrate noisy sensory input and decide the correct action

Question 3

2nd level - algorithm summary

Answer 3

What abstract method of processing information could you use to solve this problem?
Drift-diffusion model of evidence accumulation
- Model explains
- Frequency distribution of reaction times
- Speed-accuracy trade-of

Question 4

3rd level - physical implementation summary

Answer 4

How can you implement the algorithm to solve the problem on actual hardware?
Ramping up neural activity to a threshold
- Some evidence but neural mechanisms not completely understood

Question 5

What is the link between code breaking and sensory decision breaking?

Answer 5

Brain decodes message into appropriate behaviour

Code-breaking and sensory decision making can use the same algorithm with different physical implementations

Question 6

Why did decision theory helped crack the Enigma code?

Answer 6

Primitive computer that could test rotor combinations
In depth means the messages were coded by the identical rotor configuration
If you knew the offsets to reach in depth for several pairs of messages, you could guess the first rotor

Question 7

How does decision theory reveal if two messages are in depth?

Answer 7

In language, some letters occur more often than others
Therefore if you overlay two texts, you will get overlapping letters more often than chance
- Expect 1/13 chance of overlaps

Question 8

What is a Banbury sheet?

Answer 8

Slide Banbury sheets across each other looking for matches and mismatches
If holes aligned it would show matching letters

Question 9

What is the Banburismus strategy for evidence accumulation?

Answer 9

Every time you have a match the evidence favours the two texts are in depth
Every time you have a mismatch goes against the evidence
If you have matches at a rate higher than 1/26 – at some point youd reach the threshold of evidence needed
- Can then conclude if in depth or not
- Provides a natural stopping point once enough evidence has been collected

Question 10

Stimulation of Banburismus

Answer 10

Normally after around 300 characters a decision can be made
- Eventually the ups outweigh the downs so you hit the decision bound threshold
Decisions that take longer are just due to chance and the organisation of the letters

Question 11

What is the drift-diffusion model?

Answer 11

Move up or down depending on the sensory evidence you have found
- Eventually you will reach a threshold
- Helps you decide if it is a match or not
Evidence accumulation is noisy so a distribution of reaction times

Question 12

What is a typical experiment to test the drift-diffusion model?

Answer 12

1. Fixation of dot on screen
2. Targets appear
3. Moving dots appear - have to decide which direction the dots are moving
4. If moving to the right – the subjects eyes move to the right

Question 13

What is the basis of the drift-diffusion model?

Answer 13

The drift-diffusion model comes from the distribution reaction times

Question 14

The distribution of reaction times is?

Answer 14

Non-Gaussian – skewed right

Question 15

How can the Gaussian distribution of reaction times be recovered?

Answer 15

If you scale x-axis according to reciprocal of reaction time (1/RT)

• Explained if underlying variable of Gaussian distribution is the rate of evidence accumulation
• Rate of evidnece distribution is normally distributed
• This is translanted into a distribution of reaction times which is skewed to the right

Question 16

How do you calculate the rate of evidence accumulation?

Answer 16

Height/width of line
Height – amount of evidence
Width – reaction time

Question 17

Reaction times in Drosophila experiment

Answer 17

Flies in a chamber can choose between a bad odour and lower concentration of the bad odour

• Easy task - odour vs. 0.1 concentration
• Hard task - odour vs. 0.9 concentration

Question 18

What distribution does the reaction times in Drosophila show?

Answer 18

Rightward skew
Easy task most rightward
Central feature of sensory decision making

Question 19

What experiment is used to test speed-accuracy trade-off?

Answer 19

1. Mouse pokes its nose in the odour port
2. When it hears a buzz, it has to stop sniffing and choose a water port based on which odour it smelled
3. Reward only for going to the correct water port

Question 20

What happens if a mouse’s odour sampling is cut short by the buzzer?

Answer 20

Makes less accurate decisions

Question 21

How does the drift-diffusion model explain why being forced to decide too early makes decisions less accurate?

Answer 21

Moving the decision bound adjusts trade-off between speed and accuracy
Closer together decision bounds – less accurate but decision made quicker

Question 22

How are decision bounds flexible?

Answer 22

Decision bounds get lower over the course of decision making as the need

Question 23

Where is evidence accumulated in the brain - recordings from lateral intraparietal neurons of monkeys

Answer 23

Responsible for generating eye sacades
Increase in spike rate of neurons in visual cortex when evidence is stronger
- Stronger sensory input – LIP neuron activity changes faster
- Line up LIP neuron activity with animals behaviour – LIP neuron reaches common threshold - 70 ms before action

Question 24

Where is evidence accumulated in the brain - dendritic integration in Drosophila Kenyon cells

Answer 24

Kenyon cells store olfactory memories

A fly mutant, FoxP, takes abnormally long to decide between two odours

Question 25

Why do FoxP mutants take longer to decide?

Answer 25

FoxP mutant Kenyon cells have excess K+ channel expression making them leaky
- Normally FoxP transcription factor represses this K channel expression
Slows down their rate of evidence accumulation
- Takes Kenyon cell depolarisation a lot longer
- Depolarisation up to spike threshold