I am facing a problem and do not know whether it is even solvable: I want to predict the behaviour of a system using a DNN, say a CNN, in the sense that I want to predict the time and intensity of a maneuver performed by a player. Let's leave it relatively abstract like this, the details do not matter.

My question is now whether there is any way of knowing how well my CNN performs. My goal would be to derive statements of the form "With x% probability, the correct maneuver angle is within the predicted angle +-y%".

Can such statements be derived e.g. using statistical analysis of the test data? I saw approaches toward verification and validation of DNNs using satisfiability modulo theory, but did not really understand the details. Would this be applicable here? It seems a little overkill...


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.