Search code examples
neural-networkbiological-neural-network

Training neural network -

I am planning to use neural network to authenticate users. For example, the same idea is used by Coursera - they are authenticating user by his/her typing pattern.

I am going to use neural network for this. An input would be a vector of normalized and statistically clean values. Output - how likely is that this is a current user.

Before usage, training data set will be collected from the user.

However, I am worried that during training I will provide data set only for a current user. No data will be provided for recognizing incorrect user.

Maybe you can advise if it is necessary to provide incorrect user's data for training? Or maybe there are some network types / configurations better suited for this?

Thanks in advance!

Solution

  • Yes, it is necessary to provide counter-examples. If you don't, then you are training your system to always report "yes".