Ilp: a short look back and a longer look forward
The Journal of Machine Learning Research
QG/GA: a stochastic search for Progol
Machine Learning
Hi-index | 0.00 |
I have chosen the title of this talk in the same spirit as the early Buddhists, who often resorted to enumeration as a mnemonic device. (cf. the Three Jewels, the Four Noble Truths, the Eight-Fold Path containing the Five Elementary Precepts, the Six Virtues, the 84,000 Teachings and so on). Specifically, it is my intention to remind you of an invited talk at ILP 2000 by David Page, in which five areas were proposed as pressing issues for Inductive Logic Programming to address. For those who came in late, these were – Incorporating probabilities. – Novel search methods. – Techniques for parallel ILP. – Using special-purpose reasoners. – Enhancing human-computer interaction. Five years is a reasonable length of time to take stock and ask: has progress been made on each of these fronts? We know di.erent approaches have been proposed for incorporating probabilities, some unusual search methods have been developed and a few techniques for concurrent processing have been investigated. But can we do things now that we could not five years ago (or at least, could do so only with great di.culty then)? The principal hindrances to a direct answer are that there are no yardsticks established for measuring progress, nor has the same ILP system been used in all cases. As a result, we can do little more that list out who has done what and how over the past five years.