Information theoretic upper bounds on the number of distinguishable classes
Catherine M. Keller, Matthew H. Ho, Prabahan Basu, Gary H. Whipple · 2013
This paper examines data driven information theoretic upper bounds on the number of classes that can be distinguished by machine-learning classification systems as a function of the signal-to-noise ratio (SNR) of the features. Fano upper bounds are derived with desired classification error as a parameter. A simulation example is used to explore the bounds.