Download PDFOpen PDF in browserMulti-Layer Perceptron: Overcoming the Local Minima Problem: Hierarchical Binary ClassifiersEasyChair Preprint 149974 pages•Date: September 22, 2024AbstractIt is well known that the gradient descent rule employed in training the Multi-Layer Perceptron (MLP) could get stuck in a local minima of the error/loss function ( based on mean squared error ). We reason that by realizing MLP using a cascade of binary classifiers ( MLP with single neuron in the output layer ), the Hierarchical classification approach overcomes the local minima problem ( since the loss function of each binary classifier is a paraboloid ). Several innovative ideas related to such Artificial Neural Network architecture are being proposed. Keyphrases: Cascade Architecture, Classification, Local minima Problem, Multi Layer Perceptron, binary classifiers
|