Multiple choice Flashcards
Which of the following does not reflects spiking activity?
Single-unit activity (SUA)
Multi-unit activity (MUA)
Low frequency components of the EFP (extracellular field potentials)
High frequency components of the EFP
Low frequency components of the EFP (extracellular field potentials)
What needs to be taken into account when modelling the BOLD-signal?
The shape of the haemodynamic response function
Slow fluctuations in the signal
The effects of the experimental conditions
All of the above
All of the above
What is the hemodynamic response?
A way for the neuron to cool itself
A way for the neuron to rapidly get rid of waste matter
A way for the neuron to inhibit ‘firing’
A way to get more oxygen and energy to the neuron
A way to get more oxygen and energy to the neuron
What is a classifier in multivariate analysis of fMRI?
A number between 0-1 that specifies the level of activation in a voxel
An algorithm used to separate brain activation according to category
A linear model that predicts the amount of neuronal activity in a group of voxels
The voxel with the most influence on model performance.
An algorithm used to separate brain activation according to category
Which of the following statements regarding multivariate analysis is false?
Multivariate analysis measures the difference in activation patterns of voxels
Encoding models can be used to infer stimuli from brain activation in multivariate analysis
Assuming different brain regions are the same is a pitfall of multivariate analysis
Multivariate analysis measures if activation is significant on a per-voxel basis
Multivariate analysis measures if activation is significant on a per-voxel basis
What is the Curse of Dimensionality and why is it a problem for multivariate fMRI analysis?
A high number of voxels and low number of samples leads to false positives
Warping of voxel space leads to false negatives
A genetic deformity of brain regions leads to false inferences
The natural layering of the cortex leads to difficulties in fMRI measurements
A high number of voxels and low number of samples leads to false positives
What characterises the bottom-up approach?
It is the same approach as Allan Turing had on human intelligence <3
It is to move from theory to data
To investigate the brain with focus on making theory based on data
To replicate cognitive behaviour
To investigate the brain with focus on making theory based on data
Which statement is true about brain organization? Neuron are….
arranged in local clusters with fewer long distance connections
connected to all other neurons
all randomly connected across the brain to a fraction of all neurons
arranged in local clusters with random long range connections
arranged in local clusters with fewer long distance connections
Inhibitory feedback happens when….
…the excited neuron reduces the activity of its neighbors.
…the interneuron gets excited by one population of cells and inhibits a different population of cells.
…the interneuron inhibits the population of cells that it’s just been excited by.
…the neuron inhibits all other neurons in the brain
…the interneuron inhibits the population of cells that it’s just been excited by.
Slow brain oscillations
Cannot propagate
Can propagate over small areas
Can propagate over larger areas
Can propagate everywhere
Can propagate over larger areas
The presence of pink 1/f noise indicates
Oscillations at different scales are temporally linked
Oscillations at different scales are independent
Oscillations at high frequencies are less costly
Oscillations at high frequencies are more costly
Oscillations at different scales are temporally linked
Are neurons integrate and fire systems?
Definitely yes, like we see in neural networks
Definitely no, brains are not like neural networks
It was the prevailing model until the 1980s, but no longer
It has been the prevailing model since the 1980s
It was the prevailing model until the 1980s, but no longer
What exactly is meant by “shared weights” in convolutional neural networks?
Per feature map, the same weights are applied to each local receptive field
All feature maps in the network have the same weights
All feature maps in a layer have the same weights
You must initialize all weights with the same value
Per feature map, the same weights are applied to each local receptive field
Are convolutional neural networks designed to closely parallel biological vision system?
Yes, CNNs provide framework for fully biological brain-computational models
No, but the activity of CNNs can be related to activity of the visual system
No, only recurrent NNs are completely parallel to primate visual system
Yes, CNNs have proven to be 90% similar to biological brains
No, but the activity of CNNs can be related to activity of the visual system
How do you pass the output of a conv. layer to a fully connected layer?
Flatten it
Multiply every input pixel by the total number of shared weights
Calculate the gradient and subtract it from the loss
import numpy
Flatten it
