Good afternoon,
I'm having some problems concerning the question 8.4.3 of the mini-project 2. Here, we are asked to create a number of patterns K, with the first one being the checkerboard one and the other being created randomly. K is found by using C_store.
We have to run the code several times to notice some specific event. Now, what I don't understand is that what I expected to see was to notice sometimes the appearance of one wrong pixel after the first iteration (which then stabilizes). However, in some cases, I saw that this wrong pixel "spreads" by creating new wrong pixels in the following steps of the dynamics.
By looking at the book, in particular at the grey box called "Erroneous bits" in part 17.2.4, it seems that this phenomenon of "avalanche" of wrong pixels should be avoided if we use a C_store = 0.138 ... But it does not look to be the case. Why is it? I'm also wondering if it's likely to be due to the size of the network... indeed this rule is done by considering a network with N which goes to infinity (and indeed in the book is 10'000), while we have only 100 neurons.
Thank you in advance,
Leonardo Pollina
