The dataset contains the data which are the outputs from an unknown function 
$$ y=f(x\_1,x\_2). $$
Note that the given training dataset has been disturbed by some noise. You are now required to design a network to remove the introduced noise. After finishing your network training, you need to produce the outputs for a given testing dataset. Kaggle will compute the **mean square error** (MSE) between your outputs and the true outputs for the performance evaluation of your network.