Theorems on Positive Data: On the Uniqueness of NMF

Figure 4

The graph shows the connection between the norm of the additive error ‖𝐍‖_𝐹 and the estimation error of the underlying model 𝐽_(𝐖,𝐇)(𝐖^′,𝐇^′). The two noise matrices from Example 4, 𝐍_𝑁 and 𝐍_𝑀, are plotted. In this example, the curves are aligned for small errors, and for larger errors, the model error 𝐍_𝑀 results in much larger estimation errors.