I am reviewing my Neural Network lectures and I have a doubt: My book's (Haykin) batch PTA describes a cost function which is defined over the set of the misclassified inputs.

I have always been taught to use MSE < X as a stopping condition for the training process. Is the batch case different? Should I use as stopping condition size(misclassified) < Y (and as a consequence when the weight change is very little)?

Moreover, the book uses the same symbol for both the training set and the misclassified input set. Does this mean that my training set changes each epoch?

  • $\begingroup$ What does "MSE < X" actually mean? I can infer that MSE is the mean-squared error and X is maybe the input, but what kind of operation is that? Less than? It is not clear what your stopping condition is. What is the name of the book? To prevent overfitting, you can often detect it by seeing an improvement in the performance (e.g. accuracy) on the training dataset, but a decrease in performance on the evaluation dataset. This is often a sign of overfitting. See: stats.stackexchange.com/q/131233/82135. $\endgroup$
    – nbro
    Feb 6, 2019 at 21:09
  • $\begingroup$ Sorry for the lack of clarity. for "MSE < X" I meant "MSE lower than a certain value X". I always been taught to use this stopping condition to prevent overfitting (as you told me) and stop the training when a certain precision on the train set (let's say 95%) is met. $\endgroup$ Feb 8, 2019 at 0:18


You must log in to answer this question.

Browse other questions tagged .