Introducing the Special Issue of Machine Learning Selected from Papers Presented at the 1997 Conference on Computational Learning Theory, COLT’97

The aim of the COLT special issues is to provide a forum for COLT conference papers which though theoretical have been judged to have significant practical ramifications and hence to be of potential interest to a broader spectrum of Machine Learning readers. The papers included in this special issue cover a broad range of topics, their practical significance being in some cases of the form ‘how to do it’, while in others it is of the form ‘this could be hard’. The paper most clearly aiming at a particular practical application is by Blum and Kalai entitled,‘Universal portfolios with and without transaction costs’. The paper considers the problem of rebalancing an investment portfolio on-line in such a way that the result is competitive with the best constant rebalancing determined in hindsight. They extend Cover’s universal algorithm to take into account transaction costs, and at the same time give a randomized …