What should i do if my validation and test error are reducing, as the number of iterations are increasing?

Question

I built my first neural network from scratch. The results I get are The number of iterations is 10 The train error is: 0.899150924069 The validation error is: 0.897199100249 The test error is: 0.871916585165 The number of iterations is 1000 The train error is: 0.832524570025 The validation error is: 0.832585995086 The test error is: 0.842663227075 What results should i report? I use k-fold cross validation and these are the average of my results

Answer 1

Your validation and test error are supposed to reduce while training. This means you're network is 'learning'. However, you should give your neural network a goal: e.g. a MSE error of 0.05. Once you have reached this goal, report the error and the number of iterations as a result!

PS: ~0.8 is a really high error!