Introduction to Computational Neuroscience P2 Flashcards
define neural network
a network of neurons
each neuron is connected to other neurons through synapses
each synapse has a weight (parameter)
what type of network is an artificial neural network
multi layer perceptron
(maps input onto output)
components of MLP
-input
-hidden layers
-output
non linear relationship
MLP features
abstract model
not accurate of real neuronal circuits
feedforward (unlike brain which is feedback)
what is a more complex neural network
convolutional neural networks CNN
similar to visual system
what does a feature map show in CNN (visual systems)
low level feature (lines)
mid level feature
high level feature (faces)
what does fMRI of the CNN show
higher level layers in the CNN associated with higher level components of visual system
uses of computational models
physical neural function and algorithms that guide human or animal behaviour (mental health problems)
which model shows the model of learning
rescorla wagner model
role of rescorla wagner model
uses prediction errors to update its expectations
what is a prediction error
difference between what was expected and what was observed
what parameter is used in the rescorla wagner model
a, learning rate (how fast you learn from prediction errors)
rescorla wagner formula
Vt+1 = Vt + a(outcome-Vt)
Vt+1 = updated value expectation
Vt = current value expectation
a = learning rate 0-1
outcome-Vt = prediction error
how does ‘a’ change
lower a = less fluctuations, low learning rate
higher a = more fluctuations,environment is variable
roughly correct learning rate
a = 0.4-0.5
