Hello everybody,
I am ambiguous about a learning problem.
I am working on a task of anomaly detection. I train a model of only normality patterns, then classify a new sample by its likelihood (or similar quantities). I determine several hyperparameters (e.g. number of training epochs) by cross-validation, in which each evaluation set contains samples of normality and abnormality.
So I want to know your idea about this learning.
Thank you!
Hi… I am working on anomaly detection problem too, with similar algo as you have mentioned. In my opinion this category belongs to One Class classification. Hence I feel it’s semi-supervised, because algo is not forced to find the structure in the data, rather we are providing half of the classes.
Hello @rkmalaiya, thanks for the link.
Could you tell me if you implemented one class classification using deep learning\fast.ai specifically?
I would love to get some resource recommendation on that.