Homework and notes for Udacity Deep learning courses
###Basic concept
One-Hot encoding In multi-class classification problem, encode the true class 1 and all else 0. Same as one-vs-all encoding
Cross-entropy D(S,L) = -Sum(S*log(L)) #L stands for labels and S stands for distribution D(S,L) != D(L,S) not symmetric
Feature scaling for images. Since pixels have a range (0,255), use (value-128)/128 to scale them.
ReLU Rrectified linear unit. It is the activation function. Defined as: h=max(0,a) where a=Wx+b What is ReLU
Compare to sigmoid function, two additional major benefits of ReLUs are sparsity and a reduced likelihood of vanishing gradient.
-
One major benefit is the reduced likelihood of the gradient to vanish. This arises when a>0. In this regime the gradient has a constant value. In contrast, the gradient of sigmoids becomes increasingly small as the absolute value of x increases. The constant gradient of ReLUs results in faster learning.
-
The other benefit of ReLUs is sparsity. Sparsity arises when a≤0. The more such units that exist in a layer the more sparse the resulting representation. Sigmoids on the other hand are always likely to generate some non-zero value resulting in dense representations. Sparse representations seem to be more beneficial than dense representations.
**Dropout
https://www.quora.com/How-does-the-dropout-method-work-in-deep-learning
Compare two images Very detailed tech from StackOverflow
Intercept of logistic regerssion About interpretation about fit intercept or not
CNN
A good example to understand tensorflow CNN, also a good blog to follow