An unfeasability view of neural network learning
arXiv:2201.00945v1
AI Analysis
This provides a theoretical limitation for neural network learning, showing incremental progress in understanding feasibility constraints.
The paper tackles the problem of perfect learning in multilayer neural networks by proving that continuously differentiable perfect learning algorithms do not exist when the dataset size exceeds the number of parameters and activation functions are logistic, tanh, or sin.
We define the notion of a continuously differentiable perfect learning algorithm for multilayer neural network architectures and show that such algorithms don't exist provided that the length of the data set exceeds the number of involved parameters and the activation functions are logistic, tanh or sin.