Early applications of machine learning

Terry Benzschawel

Neural network practitioners face numerous criticisms. Many of these have proven unfounded and most are possible to overcome. Nevertheless, these criticisms have contributed, at least in part, to the slow adoption of machine learning applications in banking and finance in particular.

A common criticism is that neural networks require large amounts of data for accurate predictions. This is true, as there are more parameters (ie, weights) to determine. However, neural networks can capture complex relationships between input variables, and data is necessary to reveal these. Furthermore, the availability of larger data sets has become common in recent years and will continue to increase, thereby necessitating techniques designed to make use of that information.

Neural networks are also criticised for being computationally intensive and difficult to implement. There is some validity to this in that greater computing power is required as data sets become large, but technology has advanced rapidly (with the invention of, for example, graphics processing cards and quantum computing). Given such advances in technology and the availability of large data sets, machine learning methods continue