RESUMEN

Authors have performed learning algorithms, based on mixtures of experts, who achieve a good performance under severe time/cost restrictions, and that can be applied to non-stationary data. This is of particular interest for applications like quality of Service (QoS) prediction on IP data networks (see [12]). In this paper we show how can all the properties of this algorithms be proved in a strictly rigorous manner, with no other tools that elementary martingale theory at hand.

RESUMEN

If you plan to license a piece of code to a company, what objectively testable guarantees could you offer about the source code so that the people who will be responsible for its future development feel comfortable taking on the challenge? What criteria collectively constitute a definitive, objective trademark of good software? Hakan and a colleague analyze this question in detail.