Visualization of neural networks using saliency maps
Niels J.S. Mørch, Ulrik Kjems, Lars Kai Hansen, Claus Svarer, Ian G. Law, Benny E. Lautrup, Stephen C. Strother, Kelly Rehm · 2002
The saliency map is proposed as a new method for understanding and visualizing the nonlinearities embedded in feedforward neural networks, with emphasis on the ill-posed case, where the dimensionality of the input-field by far exceeds the number of examples. Several levels of approximations are discussed. The saliency maps are applied to medical imaging (PET-scans) for identification of paradigm-relevant regions in the human brain.