Machine Learning idea – Small input large recall memory loss. Hack the loss with .array for two zero loss values. One MSE and one 1-Recall 50%cifar10 loss = F.mean_squared_error(y,yt) rec = F.cast(F.recall(y2,F.argmax(yt2,axis=1))[0],np.float32) loss.array = F.mean(1-rec).array

Machine Learning idea – Small input large recall memory loss. Hack the loss with .array for two zero loss values.

One MSE and one 1-Recall 50% @ cifar10

loss = F.mean_squared_error(y,yt) rec = F.cast(F.recall(y2,F.argmax(yt2,axis=1))[0],np.float32) loss.array = F.mean(1-rec).array