It's interesting how pervasive this type of pyramid structure is in computer vision / computer graphics. For example, Gaussian pyramids and Laplacian pyramids are used a lot in computational photography: https://en.wikipedia.org/wiki/Pyramid_(image_processing)#Gaussian_pyramid
Laplacian pyramids can be used to analyze the different frequencies within an image.
I wonder whether logarithmically-spaced mipmaps lead to perceptibly excessive blurring. While logarithmically-spaced mipmaps guarantee constant memory overhead (as shown in the next slide) while, for example, linearly-spaced mipmaps would cause linear memory overhead, they would also do a better job of filtering out only details finer than the nyquist frequency.