Differences between online and batch learning

I can not understand the difference between online and batch learning, could someone throw some light on this topic?

As an example of batch we have the SVM, Logistic Regression, Naive Bayes:

for 1 ... N
    w <- update(T, w)
return w

T is the whole training dataset with the inputs and their labels.

While online learning as the perceptron would be something like this:

for 1 ... N
   for (x_t, y_t) ∈ T
       w <- update ((x_t,y_t); w)
   end for
return w

I imagine that the difference is the way to update the weights, but I'm not sure.

Any advice?

Thank you!