Falsification and future performance


We information-theoretically reformulate two measures of capacity from statistical learning theory: empirical VC-entropy and empirical Rademacher complexity. We show these capacity measures count the number of hypotheses about a dataset that a learning algorithm falsifies when it finds the classifier in its repertoire minimizing empirical risk. It then follows from that the future performance of predictors on unseen data is controlled in part by how many hypotheses the learner falsifies. As a corollary we show that empirical VC-entropy quantifies the message length of the true hypothesis in the optimal code of a particular probability distribution, the so-called actual repertoire.



External links

Setup an account with your affiliations in order to access resources via your University's proxy server

Through your library

  • Only published works are available at libraries.

Similar books and articles

Systems, inquiry, and the meanings of falsification.Ian I. Mitroff - 1973 - Philosophy of Science 40 (2):255-276.
Die Falsifikation Statistischer Hypothesen/The falsification of statistical hypotheses.Max Albert - 1992 - Journal for General Philosophy of Science / Zeitschrift für Allgemeine Wissenschaftstheorie 23 (1):1-32.


Added to PP

462 (#39,330)

6 months
56 (#73,473)

Historical graph of downloads
How can I increase my downloads?

Author's Profile

David Balduzzi
University of Zürich

References found in this work

On the Plurality of Worlds.David K. Lewis - 1986 - Malden, Mass.: Wiley-Blackwell.
The Logic of Scientific Discovery.K. Popper - 1959 - British Journal for the Philosophy of Science 10 (37):55-57.
On The Plurality of Worlds.Graeme Forbes - 1988 - Philosophical Quarterly 38 (151):222-240.

Add more references