Radford Neal's Research: Evaluation of Learning Methods

Many methods for "learning" relationships from data have been proposed, some based on new ideas such as neural networks, other building on traditional statistical methods. Good empirical comparisons of these methods are scarce. I was involved in the DELVE project, which hoped to remedy this situation by providing a good collection of datasets for testing learning procedures, a well-designed environment in which valid tests can be conducted, and an archive of test results for many methods. This project is now largely in abayence.

The following paper demonstrates how DELVE can be used to address interesting research questions:

Neal, R. M. (1998) ``Assessing relevance determination methods using DELVE'', in C. M. Bishop (editor), Neural Networks and Machine Learning, pp. 97-129, Springer-Verlag: abstract, associated references, postscript, pdf.


Back to Radford Neal's research interests
Back to Radford Neal's home page