Computational modelling

Question 1

What is computational about neuroscience?

Answer 1

> computer based modelling of neuronal functions at different levels
we try and understand the complex interactions in dynamic systems

Question 2

David Marr’s level of analysis:

Answer 2

1) What is the problem being solved?
or how is the brain trying to solve this problem?

2) Which algorithms can be used to solve it?
3) How are these implemented in neuronal systems? How is it being solved?

Question 3

is Marr’s level of analysis a linear model?

Answer 3

Although David Marr thought it to be - it is not

John O’Keefe figured out place cells without going through these stages

Question 4

‘Humans generalise effortlessly from limited data sets’

Which study proved this?

Answer 4

Level 1: how is the brain solving this problem? How is the brain generalise language?
Lany & Saffran (2010) – infants were ‘trained’ in an artificial language in which word categories were reliably distinguished using statistical cues

Question 5

Lany & Saffran Study illustrated:

Answer 5

–> Infants learn language patterns through statistical pairing
Infants attribute meanings of words based on statistical cues (during training)
They then categorised and generalised the pattern to include novel word pairings

Question 6

Using an algorithm (Level 2) to help us understand how to categorise

Answer 6

Bayesian Statistics
1) Prior Beliefs:
Knowledge of word categories
Generative model

2)Data
Empirical observations
We make assumptions based on prior beliefs then over time we collect data and alter these assumptions

Question 7

What is Bayesian Probability?

Answer 7

A theory in which evidence about the world is expressed in terms of ‘degrees of belief’

Question 8

How does Bayesian learning occur in the sensorimotor system?

Answer 8

Kording & Wolpert

In the most uncertain condition we are least likely to update and change, and maintain based on our prior belief

Question 9

Algorithm based on bayesian probability

Answer 9

Bayes Category Formation
Prior Belief + On-going Information
==> Effortless Categorisation

Question 10

Alogorithm for hierarchical learning:

Answer 10

Kemp & Tenenbaum
Their model brings structure learning methods closer to human abilities and may lead to a deeper computational understanding of cognitive development.

Question 11

Algorithm for bird songs:

Answer 11

Making a Generative model

->Different levels are distinguished in their temporal scales

Question 12

Step 3: Implementation

=>Dehaene-Lambertz et al. 2006

Answer 12

Sequential distribution reveal the language hierarchical processing scheme

Question 13

Which are of the brain was activated during presentation of sentences in preverbal infants?

Answer 13

Perisylvian Activation
- Adult like activation
>Dehaene-Lambertz et al. 2006

Question 14

What were the adult like responses in language processing in preverbal infants?
Dehaene-Lambertz et al. 2006

Answer 14

Fastest Responses:
- Heschl’s gyrus
Slower responses (higher cortical areas; processing happens later)
- Posterior STG
- Temporal poles
- IFG (Broca)

Question 15

When was Broca’s area activated in preverbal infants?

Answer 15

Activation in Broca after repetition of sentence

Broca Activated before babbling: activity in this region is not the consequence of sophisticated motor learning but, on the contrary, that this region may drive, through interactions with the perceptual system, the learning of the complex motor sequences required for future speech production.

Question 16

Step 3: Implementation

study by: Iglesias et al. 2013

Answer 16

Hierarchical signals of Prediction Error are processed in neuroanatomically (and physiologically) distinct regions

Question 17

Where are ‘low level’ prediction errors processed?

= Iglesias et al. 2013

Answer 17

low-level PEs about visual stimulus outcome => widespread activity in visual and supramodal areas + midbrain

Question 18

Where are ‘high level’ (abstract) prediction errors processed?
= Iglesias et al. 2013

Answer 18

high-level, abstract, PEs about stimulus probabilities in the basal forebrain.

Question 19

Step 3: Implementation

study by: Emberson et al. 2015

Answer 19

Top Down Modulation in the infant Brain - does it occur?
Is expectation-related feedback an inherent property of the cortex (i.e., operational early in development) or does expectation-related feedback develop with extensive experience and neural maturation

Question 20

What did Emberson et al. 2015 study tell us about ‘top down modulation’ in the infant brain?

Answer 20

The occipital cortex of 6-month-old infants exhibits the signature of expectation-based feedback
- this region does not respond to auditory stimuli if they are not predictive of a visual event.

These findings suggest that the young infant’s brain is already capable of some rudimentary form of expectation-based feedback

Question 21

Computational Neuroscience has helped us understand:

Answer 21

1) Top down processing in infancy
we do not learn about the world only in a bottom-up manner
We need experience with the world (bottom-up) + top-down in synchrony

2) Hierarchical learning that you experience aids your learning – learning complexity requires hierarchical interactions
3) Development is regionally heterogeneous

Question 22

Dynamic causal modelling

Cooray et al. 2016

Answer 22

MMN - cortical network develops in adolescence up to adulthood

== computational analysis can extract relevant information from neuronal measurements