Sebag, M. 2 nd order understandability of disjunctive version spaces. In Workshop on Machine Learning and Comprehensibility, IJCAI-95. Report LRI, Universit'e Paris-Sud, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....we introduced the category of notevents of evolution [161] Not events typically generate offspring that have same fitness as their parents. For the sake of convenience, we used the machine learning shell developed in the Laboratoire de M ecanique des Solides a l Ecole Polytechnique [157, 158]. Very likely, any other learning algorithm could have been used, provided that it can deal with noise 23 . 7.4 Memory based strategies of control The memory of evolution can conveniently be viewed as a set of schemas in the operator space, labeled as fatal errors, not events or success. Let M ....

M. Sebag. 2 nd order understandability of disjunctive version spaces. In Workshop on Machine Learning and Comprehensibility, IJCAI-95. Report LRI, Universit'e Paris-Sud, 1995.

....(ae entails CS) where TC = tc( X)ae. Small learning can then be viewed as a lazy, on fly learning: it produces CS, which can be viewed as a computational characterization of G and does not need to be solved in order to classify unseen example. Furthermore this classification can be justified [29]. 3.8 Complexity The number of constrained variables in CS is upper bounded by the number NoP of predicates, plus the number NoX of initial and relational variables grounded by . The number of constraints in CS is upper bounded by the number of negative substitutions NoL. If J denotes an upper ....

....a given example. In contrast, many learners are concerned with finding specific generalizations [18, 24, 20] or one discriminant generalization of an example [22, 3, 19] which can be transformed into an element of G via a post pruning phase. The issue of redundant learning was motivated in [29]: redundant learning ensures robust learning when dealing with noisy data, or data that imperfectly encode the underlying problem. The point of noise can be addressed via MDL based heuristics, as in FOIL [23] but these heuristics apply for ground examples only. The latter point makes any ....

[Article contains additional citation context not shown here]

M. Sebag. 2 nd order understandability of disjunctive version spaces. In Ws on ML and Comprehensibility, IJCAI-95. Report LRI, Universit'e Paris-Sud, 1995.

....Second, in spite of the unintelligibility of H, the classification process is definitely not a black box: For any instance E, an intelligible and concise excerpt of H that explains the classification of E can be determined 7 . More details on this second order intelligibility can be found in (Sebag 1995). 5.3 PERSPECTIVES As mentioned earlier on, on going experiments are concerned with an internal determination of optimal values of the biases (parameters) and M . Another perspective of this approach consists in using the experimental information given by the DiVS theory. Some training ....

Sebag, M. 2 nd order understandability of disjunctive version spaces. In Workshop on Machine Learning and Comprehensibility, IJCAI-95. Report LRI, Universit'e Paris-Sud, 1995.

....ae. Then, H j (E; F ) OE K E 0 ) 9 2 Sigma K (E 0 ) 8oe 2 Sigma j ; 2 H( oe) The complexity of checking whether E 0 is K subsumed by H j (E; F ) thus is O(K Theta j Theta V 2 ) The fact that DiVS involves poorly understandable hypotheses, but is in no way a black box [14], and the validation of STILL on the mutagenesis dataset [17] have been discussed elsewhere. We focus here on the neighborhood relationship defined by: E 0 is neighbor of E if E 0 is subsumed by hypothesis H(E) section 2.3) This definition is limited in two ways: it is only defined for E ....

M. Sebag. 2 nd order understandability of disjunctive version spaces. In Workshop on Machine Learning and Comprehensibility, IJCAI-95. Report LRI, Universit'e ParisSud, 1995.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC