As the paper metioned, the objective function of the model is to make the transformed input data samples to be maximally correlated under the canonical correlation analysis (CCA) loss, and in the model_graph.py you use mean_squared_error as the loss function. Is this different?
In the paper Fig.3 , the obsvered electrical current measurements need to processed by DNN F2 then calculate the cca loss, but in the model_graph.py, the label is directly fed into the mean_squared_error (Line 164). Is this different?
Hi: