Figure 7
From: Learning without loss
Time series of \(\mathtt {RRR\_err}\) in two runs of online NMF for the letter-montage data set. RRR iterations are applied in blocks of 4 to batches of size 50, and there are 40 batches in the data set. In each run of 30 epochs there are altogether \(4\times 40\times 30\) iterations