Artificial intelligence and machine learning: Practical aspects of overfitting and regularization

Abstract

Neural networks can be used to fit complex models to high dimensional data. High dimensionality often obscures the fact that the model overfits the data and it often arises in the publication industry because we are usually interested in a large number of concepts; for example, a moderate thesaurus will contain thousands of concepts. In addition, the discovery of ideas, sentiments, tendencies, and context requires that our modelling algorithms be aware of many different features such as the words themselves, length of sentences (and paragraphs), word frequency counts, phrases, punctuation, number of references, and links. Overfitting can be counterbalanced by Regularization, but the latter can also cause problems. This paper attempts to clarify the concepts of "overfitting" and "regularization" using two-dimensional graphs that demonstrate over fitting and how regularization can force a smoother fit to noisy data.

Coarse-Grained Neural Network Model of the Basal Ganglia to Simulate Reinforcement Learning Tasks

“…A typical problem with detailed and complex models containing many neuronal structures and/or many parameters is the issue of overfitting [ 37 ]. A model with too little capacity cannot learn the given problem, whereas a model with too much capacity can learn it too well and overfit the training dataset; thus, such a model does not have the ability to generalize the knowledge well [ 38 ]. Thus, the selection of a model size that maximizes generalization is an important topic that has been given a lot of attention in the last years as the field of cognitive neuroscience develops [ 39 ].…”