Curse of Dimensionality: Classifying Large Multi-Dimensional Images with Neural Networks
Rudolf Hanka, Thomas P. Harte · Birkhäuser Boston eBooks · 1997
The term ‘curse of dimensionality’ is used to describe either the problems associated with the feasibility of density estimation in many dimensions, or the complexity of computations connected with certain types of signal processing. The use of Back-propagation neural networks to classify large 4-D MRI images is a typical example of the latter: the complexity of computations involved is such that it could render the use of neural networks too slow to be of practical clinical use. It is shown that the process could be speeded up by several orders of magnitude by using FFT-based convolution during the input stage of the classification. Further, hardware-dependent computational gain is possible when the FFT is replaced by a number theoretic transform such as the Fermat number transform.