Before you turn this problem in, make sure everything runs as expected. First, restart the
kernel (in the menubar, select Kernel→Restart) and then run all cells (in the menubar,
select Cell→Run All).
Make sure you fill in any place that says YOUR CODE HERE or "YOUR ANSWER HERE", as
well as your name and collaborators below:
NAME = Kiki Boumans
COLLABORATORS =
File "<ipython-input-1-342a7c538cbf>", line 1
NAME = Kiki Boumans
^
SyntaxError: invalid syntax
Assignments week 3
Complete the assignments below, save the notebook and submit them on canvas.
Assignment 3.1
In the last exercise in this week's notebook you saw that a Hopfield network that has
learned all possible patterns has a weight matrix that has ones along the diagonal and is
zero everywhere else. Such a network can sustain all possible patterns. However, that is
not actually good for memory recall!
Explain why a network that can sustain all possible patterns will not be able to perform
cued memory recall. Think, for example, about the case where the patterns in your network
are images and each unit represents a pixel that is either black or white. What would
happen if we present a noisy image to a network that can sustain all possible patterns?
Within associative memory (or memory recall) the attractor states of this network can be
seen as stored patterns. This is because a certain cue or stimulus is assigned to an object
which can cause an activation pattern of a certain stored memory. This means that an
input-stimulus generates an output-memory. The Hopfield network that has learned all
possible patterns will recognize every input and will always create an output. However,
you want the model to recognize specific input and generate the output that you are
actually looking for. Thus, stimuli of objects that are not related in a specific memory
should be rejected directly. Because the Hopfield network has a treshold of 0, it will
recognize every input. This means the Hopfield network will recognize all the stimuli,
which causes there to be no more cued memory recall, because all the stimuli are now cues
that cause activation of memory patterns.
kernel (in the menubar, select Kernel→Restart) and then run all cells (in the menubar,
select Cell→Run All).
Make sure you fill in any place that says YOUR CODE HERE or "YOUR ANSWER HERE", as
well as your name and collaborators below:
NAME = Kiki Boumans
COLLABORATORS =
File "<ipython-input-1-342a7c538cbf>", line 1
NAME = Kiki Boumans
^
SyntaxError: invalid syntax
Assignments week 3
Complete the assignments below, save the notebook and submit them on canvas.
Assignment 3.1
In the last exercise in this week's notebook you saw that a Hopfield network that has
learned all possible patterns has a weight matrix that has ones along the diagonal and is
zero everywhere else. Such a network can sustain all possible patterns. However, that is
not actually good for memory recall!
Explain why a network that can sustain all possible patterns will not be able to perform
cued memory recall. Think, for example, about the case where the patterns in your network
are images and each unit represents a pixel that is either black or white. What would
happen if we present a noisy image to a network that can sustain all possible patterns?
Within associative memory (or memory recall) the attractor states of this network can be
seen as stored patterns. This is because a certain cue or stimulus is assigned to an object
which can cause an activation pattern of a certain stored memory. This means that an
input-stimulus generates an output-memory. The Hopfield network that has learned all
possible patterns will recognize every input and will always create an output. However,
you want the model to recognize specific input and generate the output that you are
actually looking for. Thus, stimuli of objects that are not related in a specific memory
should be rejected directly. Because the Hopfield network has a treshold of 0, it will
recognize every input. This means the Hopfield network will recognize all the stimuli,
which causes there to be no more cued memory recall, because all the stimuli are now cues
that cause activation of memory patterns.
