P. T. Baffes and R. J. Mooney. Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, Chambery, France, 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... e.g. Audrey [WP93] Fonte [MB88] Either [OM94] and Delta [LDRG94] each consider adding or deleting antecedents or rules. Our analysis, and results, can easily be applied to many other types of modifications e.g. specializing or generalizing antecedents [OM94] using n of m rules [BM93], or merging rules and removing chains of rules that produced incorrect results [Coh90, Coh92] 2 While these projects provide empirical evidence for the effectiveness of their specific algorithms, and deal with classification (i.e. determining whether a given element or tuple is a member of ....

Paul T. Baffes and Raymond J. Mooney. Symbolic revision of theories with M-of-N rules. In Proceedings of IJCAI-93, August 1993.

.... learning in several medical domains [Spackman, 1988] Ting [1991] demonstrates the performance advantage of MoN as well as ID2 of 3 over C4.5rules in terms of higher prediction accuracy and smaller theory size in a biology domain (Splice junction) Neither, a propositional theory refinement system [Baffes and Mooney, 1993], is capable of revising M of N rules by modifying the M value of M of N representations. It is shown to be more accurate in the Promoters domain than the similar system, called Either, that can only revises conjunction and disjunction rules. 2.3.2 Other relevant constructive induction ....

....X of N representations. When building a decision tree, all of them construct one new attribute for each decision node using the local training set. Instead of building decision trees, Crls [Spackman, 1988] and MoN [Ting, 1994] learn M of N rules. The symbolic theory revision system Neither [Baffes and Mooney, 1993] refines M of N rules. The rule learning algorithms Induce [Michalski, 1978] AQ17 dci, and AQ17 mci [Bloedorn et al. 1993] use the counting operator 22 #VarEQ(x) to construct new attributes that count the number of attributes which take the value x. For primitive boolean attributes, a boolean ....

P.T. Baffes and R.J. Mooney, Symbolic revision of theories with M-of-N rules. Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, San Mateo, CA: Morgan Kaufmann, 1135-1140.

....send an ID5 decision tree, to another agent, which could use it, but unless it uses the ID5 algorithm, it could not modify the hypothesis with any examples it possesses. Theory Revision The next type of learning algorithm that might be used is a theory revision algorithm, for example NEITHER (Baffes and Mooney, 1993). Designed originally to perform minor revision to expert rules, theory revision systems search the hypothesis space close to the input hypothesis. Typically, they can delete or add antecedents, add or delete clauses. However, they do rely on the heuristic that the correct revision is close to ....

....the test set, and allocating the remaining folds to the appropriate agents. Each k fold experiment was also repeated ten times, using a different random k fold of the examples (but using the same shuffle for both theory revision and version space using agents) In both experiments, we use NEITHER (Baffes Mooney, 1993) theory revision algorithm. It should be noted that we are not evaluating this specific theory revision algorithm s performance, or indeed in general) Theory revision is generally accepted as making small modifications to an approximately correct theory usually supplied by a human expert. We are ....

P. T. Baffes and R. J. Mooney (1993). Symbolic Revision of Theories With M-of-N Rules, in Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence (IJCAI-93), pages 1135-1140, Chambery, France.

....series of empirical analyses of the S NG approach, using problems from the UCI machine learning repository (Merz Murphy, 1998) 1. Complete the comparison of the multi agent model with version spaces constructed using Espresso, against the multi agent model with theory revision using NEITHER (Baffes and Mooney, 1993). Compare these with the performance of the FOIL learning algorithm with the appropriate percentage of the examples. Both accuracy and time to learn will be measured. In particular we plan to measure worst and average case behaviour. Worst case behaviour is important as theory revision methods can ....

P. T. Baffes and R. J. Mooney (1993). Symbolic Revision of Theories With M-of-N Rules, in Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence (IJCAI-93), pages 1135-1140, Chambery, France.

....Theory revision systems have evolved in a Machine Learning background where Horn Clauses are a common representation. Therefore a main aim of researchers in theory revision has been extend the techniques from propositional theories (e.g. EITHER) through M of N theories (e.g. NEITHER (Baffes Mooney, 1993)) 3 , to restricted first order theories systems (e.g. AUDREY II, FORTE, WHY, MOBAL, CLINT) In contrast many expert systems are propositional, or at least use representational schema which do not need the full power of First Order Predicate Calculus. The inference engine is an integral part of ....

Baffes, P. T., & Mooney, R. J. (1993). Symbolic revision of theories with M-of-N rules.

....5 Knowledge Refinement As research continues on the problem of using ML for knowledge acquisition, we will develop more guided approaches than the weak search methods. One step that has already been taken in this direction is that of automatically refining incorrect or partial domain knowledge [4, 11, 12, 15, 16, 20, 21, 22, 26, 30, 31, 32, 33, 34, 47, 50]. Even if we do not have a fully satisfactory set of rules for solving a problem, our learning algorithms can still benefit from the incomplete knowledge we do have. Knowledge refinement systems such as those cited are often able to use partial knowledge to produce better solutions to real world ....

....systems, like other learning systems, can be symbolic or connectionist. A symbolic approach typically starts with a set of imperfect rules from a human expert and iteratively modifies it in order to improve its correctness or coverage, e.g. by adding and deleting terms. Either [32, 33] and Neither [4] are systems which refine propositional Horn clause rule sets in such a manner, and Forte [26] extends the technique to function free Horn clause representations of logic programs. The Kbann family of algorithms represents a connectionist knowledge refinement approach. It translates a set of ....

P.T. Baffes and R.J. Mooney. Symbolic revision of theories with M-of-N rules. In Proc. 13th Int'l Joint Conf. on Artificial Intelligence, pages 1135--1140, Chamb'ery, Savoie, France, 1993. Morgan Kaufmann.

....Towell, Shavlik, Noordewier, 1990; Pazzani Kibler, 1992) have sought ways of automating the process of fine tuning a rule base. Throughout this document, I refer to this process as rule base revision, though other authors have called it theory revision (Mitchell, Keller, Kedar Cabelli, 1986; Baffes Mooney, 1992; Cain, 1991; Ginsberg, 1990), or theory refinement (Buntine, 1991; Ourston Mooney, 1994) I prefer the former in an attempt to avoid any confusion with the scientific usage of the word theory. Given a set of rules, the rule base revision task is to rapidly revise these rules in a way that improves their performance on a ....

Baffes, P., & Mooney, R. (1993a). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, pp. 1135--1140 Chambery, France.

.... property is the M of N concept (at least M out of N properties of a certain kind are present in an object) Problems of this type occur in various real world problems, for example, in medicine (Spackman, 1988) planning (Callan Utgoff, 1991) game playing (Fawcett Utgoff, 1991) biology (Baffes Mooney, 1993) and biochemistry (Towell Shavlik, 1994) The proposed method addresses a class of problems that require learning descriptions combining one or more M of N concepts with one or more DNF expressions. Such combined descriptions are called conditional M of N rules. The well known M of N rules are ....

....Michalski s (1983) counting arguments generalization rule to expand the original representation space. Callan and Utgoff (1991) developed a restricted form of the counting arguments rule to create a numeric function from a Boolean expression that begins with a universal quantifier. More recently, Baffes Mooney (1993) introduced the NEITHER M ofN system, which refined M of N rules by increasing or decreasing either M or N. The method described here aims at learning descriptions that combine M of N concepts with DNF expressions. To this end, it searches for attribute symmetry that is evidenced by ....

Baffes, P.T. & Mooney, R.J. (1993). Symbolic Revision of Theories with M-of-N Rules. In Proceedings of the 2nd International Workshop on Multistrategy Learning (pp. 69-75). Harpers Ferry, WV: Morgan Kaufmann.

....recursive reasoning. 3.3.4 Transformational Model Induction in ASSERT Unlike synthetic SM induction, transformational SM induction (PIXIE, SMS1, Hoppe s) has no direct counterpart in inductive ML, although theory revision comes close. For example, using the propositional theory reviser NEITHER (Baffes Mooney, 1993), ASSERT (Baffes Mooney, 1996) induces an SM from the ideal model and a student s answers to a set of multiple choice questions, and it is the SM, rather than the student behavior, that is examined for misconceptions. Table 5 shows the basic procedure of model induction in ASSERT. Like THEMIS, ....

Baffes P., & Mooney, R. (1993). Symbolic revision of theories with m-of-n rules. In Proc. International Joint Conference on Artificial Intelligence '93.

....with external processes or other agents (E9) Finally, the learning is assumed to be correct and is generally irreversible. Therefore a domain with processes that evolve would be unlearnable (E10) EITHER NEITHER, FOIL FOCL and TRAIL EITHER [Ourston and Mooney, 1990] and the more recent NEITHER [Baffes and Mooney, 1993], correct errors in Horn clause propositional logic domain theories. EITHER corrects errors in the theory as a whole. The theory s preconditions are antecedents of the Horn clauses that lead to an example being classified as belonging to the theory. The antecedents are represented in disjunctive ....

Paul T. Baffes and Raymond J. Mooney. Symbolic revision of theories with m-of-n rules. In Proceedings of the International Joint Conference on Artificial Intelligence, pages 1135--1140, 1993.

.... CRLS learns Mof N rules by employing non equivalence symmetry bias and criteria tables (Spackman, 1988) ID 2 of 3 incorporates M of N tests in decision trees (Murphy Pazzani, 1991) AQ17 DCI (Bloedorn Michalski, 1991) employs a variety of operators to construct new attributes, NEITHER MofN (Baffes Mooney, 1993) is able to refine M of N rules by increasing or decreasing either of M or N. The idea of counting attributes and counting arguments rule is related to research on detecting symmetry in Boolean functions of many variables (Michalski, 1969; 1983) and implementation of SYM programs (Jensen, ....

Baffes, P. T. and Mooney, R. J., "Symbolic Revision of Theories with M-of-N Rules," Proceedings of the 2nd International Workshop on Multistrategy Learning, Harpers Ferry, WV, pp. 69-75, 1993.

....al. 1989) however, they do not scale well since they require very large matrices to represent large networks (Yao, 1993) Other techniques (Harp et al. 1989; 6 The relationship between connectionist theoryrefinement systems and purely symbolic ones has been extensively covered (Towell, 1991; Baffes Mooney, 1993); thus we do not discuss it here. Kitano, 1990a; Dodd, 1990) only encode the most important features of the network. These indirect encoding schemes can evolve different sets of parameters along with the network s topology and have been shown to have good scalability (Yao, 1993) Regent differs ....

Baffes, P. & Mooney, R. (1993). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, (pp. 1135--1140), Chambery, France.

....on the audiology domain show that using referees can help obtain higher accuracies than those obtained by any of the individual prediction models. In this paper, two currently active research programs in machine learning, theory revision (Flann Dietterich 1989) Mooney 1993) Ourston 1991) (Baffes Mooney 1993) (Richards Mooney 1995) Towell, Shavlik, Noordewier 1990) R. S. Michalski 1993) Cohen 1992) Bergadano Giordana 1988) and bias selection (Merz 1995) Ho, Hull, Srihari 1994) Brodley 1993) Schaffer 1993) are viewed from a single perspective. Theory revision systems make use of two ....

Baffes, P. T., and Mooney, R. J. 1993. Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence.

....was also supported by a domain expert (Ourston, 1991) The second piece of domain knowledge is that the concepts in this domain tend to take the form of M of N concepts. Some of the final rules extracted in the KBANN approach take this form. This was also made clear in the NEITHER MofN system (Baffes Mooney, 1993), which added a mechanism to handle M of N concepts to the original learning mechanism of the EITHER NEITHER system. 3. M of N Interpretations of the DNA Promoters Theory We can modify the original DNA domain theory as follows, and allow more sequences to be positively classified as promoters: a) ....

....76.5 . As shown in Figure 2 a few random guesses allow us to do much better than that. On the Informativeness of the DNA Promoter Sequences Domain Theory 4. Learning with the DNA Promoters Theory The accuracies of various systems that integrate analytical and empirical learning are around 93 (Baffes Mooney, 1993). These results are typically means computed over multiple trials with 80 85 training examples and 21 26 tests examples. Our reported accuracies of 93.4 and 97.2 are not based on such splits of training and test data. Instead, they represent the maximum accuracies (relative to the database of ....

[Article contains additional citation context not shown here]

Baffes, P. T., & Mooney, R. J. (1993). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence Chambery, France.

....is at least M out of N properties of some kind must present in an object, and the logic type condition is null. Problems of this type occur in many real world problems, for example, in medicine (Spackman, 1988) planning (Callan Utgoff, 1991) game playing (Fawcett Utgoff, 1991) biology (Baffes Mooney, 1993) and biochemistry (Towell Shavlik, 1994) The proposed method can learn DNF descriptions (decision rules) one or more counting conditions (e.g. M of N concepts) or any combination of the logic type DNF descriptions with counting conditions. Thus, it extends the class of learning problems to ....

....the percentage of permutations of variable bindings satisfying the Boolean expression. Such a function is useful because it indicates the degree to which a subgoal is satisfied in a given state. More recently, the NEITHER M of N system refined M of N rules by increasing or decreasing either M or N (Baffes Mooney, 1993). The proposed method is based on the idea of counting arguments generalization rule (Michalski, 1983) the algorithm for detecting symmetry in Boolean functions of many variables (Michalski, 1969) and its subsequent implementation in the SYM program (Jensen, 1975) This work is also related ....

Baffes, P.T. & Mooney, R.J. (1993). Symbolic Revision of Theories with M-of-N Rules. In Proceedings of the 2nd International Workshop on Multistrategy Learning (pp. 69-75). Harpers Ferry, WV: Morgan Kaufmann.

....is a Prolog literal, and corresponds to an Ordered Term in the KrustTool hierarchy; the literal s predicate name is the Ordered Term s keyword. Prolog s depth first search of clauses easily provides information for our problem graph. Therefore Prolog KBSs have KrustTool representations. Neither (Baffes Mooney 1993) extends Either s refinement process by having specialised refinement operators for m of n rules. An m of n condition contains a set of n conditions, and is defined to be true if and only if at least m of the n conditions are true. Similarly, Seek (Ginsberg 1988) refines rules in a specialised ....

Baffes, P. T. & Mooney, R. J. (1993). Symbolic revision of theories with M-of-N rules, in R. Bajcsy (ed.), Proceedings of the Thirteenth IJCAI Conference, Chambery, FRANCE, pp. 1135--1140.

.... in more accurate results than inducing a knowledge base from scratch (Ourston Mooney, 1994; Towell Shavlik, 1993) Several theory refinement systems use abduction on individual examples to locate faults in a theory and suggest repairs (Ourston Mooney, 1990, 1994; Wogulis Pazzani, 1993; Baffes Mooney, 1993, 1996; Brunk, 1996) Each of these systems use abduction in a slightly different way, but the following summarizes the basic approach. For each individual positive example that is not provable from the current theory, abduction is used to determine a set of assumptions that would allow it to be ....

Baffes, P., & Mooney, R. (1993). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, pp. 1135--1140 Chambery, France.

....11:53; p.4 5 3.2. Theory Refinement Algorithms and Systems Several theory refinement systems use abduction on individual examples to locate faults in a theory and suggest repairs (Ourston and Mooney, 1990; Ourston, 1991; Ourston and Mooney, 1994; Wogulis and Pazzani, 1993; Wogulis, 1994; Baffes and Mooney, 1993; Baffes, 1994; Baffes and Mooney, 1996; Brunk, 1996) The ways in which various forms of logical abduction can be used in revising theories is also discussed and reviewed by Dimopoulos and Kakas (1996) however, they do not discuss using abduction to generalize existing clauses by deleting ....

....examples that it covers. For example, the more negative examples that are generated when the literals corresponding to an assumption set are deleted, the more complex the resulting repair is likely to be. The Either (Ourston and Mooney, 1990; Ourston and Mooney, 1994; Ourston, 1991) and Neither (Baffes and Mooney, 1993; Baffes, 1994) theory refinement systems allow multiple assumptions in order to prove an example, preferring more specific assumptions, i.e. they employ most specific abduction (Cox and Pietrzykowski, 1987) Audrey (Wogulis, 1991) Audrey II (Wogulis and Pazzani, 1993) A3 (Wogulis, 1994) and ....

[Article contains additional citation context not shown here]

Baffes, P. and R. Mooney: 1993, `Symbolic Revision of Theories with M-of-N Rules'.

....out from the start that basic design of ASSERT is not tied to a particular theory refinement algorithm. Other theory refinement systems which differ from the one presented here could be used to provide ASSERT with different or enhanced capabilities. ASSERT uses the NEITHER algorithm (Baffes, 1994; Baffes Mooney, 1993) which is based on the EITHER theory refinement system (Ourston Mooney, 1990) EITHER was chosen as a starting point because it was the most complete symbolic theory refinement system available. NEITHER is designed to work with a propositional Horn clause knowledge representation. It takes two ....

.... (integer set through pointer) REFINEMENT BASED STUDENT MODELING 9 to modify the repair to avoid new misclassifications. The whole process is embedded in a loop which continues until all misclassified examples have been accounted for. The result is that NEITHER runs very quickly (see Baffes Mooney, 1993), giving response times that are on the order of a few seconds. This is critical to an interactive tutoring environment where feedback must be generated for the student in a timely fashion. 3.2 Overview of ASSERT Having reviewed the basics of theory refinement, we can now turn to the details of ....

Baffes, P. and Mooney, R. (1993). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial intelligence, pages 1135-1140. Chambery, France.

....experts have become comfortable with this formalism. However, results in theory revision show that the accuracy of such rule bases can be dramatically improved by mapping them to a representation that employs some form of uncertain reasoning or numerical summing of evidence (Towell et al. 1990; Baffes Mooney, 1993; Mahoney Mooney, 1993) This approach also provides a very straight forward way of biasing a Bayesian network learner with some prior knowledge. An essential component of such a theory revision approach is an efficient technique that can revise the parameters of a Bayesian network composed ....

Baffes, P., & Mooney, R. (1993). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, pp. 1135--1140 Chambery, France.

....completely automatic, and by taking advantage of existing correct domain knowledge, it is able to learn more accurate models from limited training data compared to inducing a complete model from scratch. Background on Theory Refinement For its theory refinement component, ASSERT uses NEI THER (Baffes Mooney, 1993) a successor to the EITHER system developed by Ourston Mooney (1990, 1994) NEITHER employs a propositional Horn clause knowledge representation. It takes two inputs, a propositional rule base called the theory, which is repaired using a set of input examples. The examples are lists of ....

Baffes, P. and Mooney, R. (1993). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial intelligence, pages 1135-1140. Chambery, France.

....and 2 categories. The theory provided with the data set has an initial classification accuracy of 50 . The main point of the test was to illustrate whether Neither could maintain the predictive accuracy of Either while reducing execution time. Other tests of the Neither algorithm are described in (Baffes and Mooney, 1993). The experiments proceeded as follows. Each data set was divided into training and test sets. Training sets were further divided into subsets, so that the algorithms could be evaluated with varying amounts of training data. After training, each system s accuracy was recorded on the test set. To ....

Baffes, P. and Mooney, R. (1993). Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial intelligence.

No context found.

P. T. Baffes and R. J. Mooney. Symbolic revision of theories with M-of-N rules. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence, Chambery, France, 1993.

No context found.

P. Baffes and R. Mooney, Symbolic revision of theories with m-of-n rules, Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence (IJCAI-93), Chambery, France 1135--1140.

No context found.

Baffes P., & Mooney, R. (1993). Symbolic revision of theories with m-of-n rules. In Proc. International Joint Conference on Artificial Intelligence '93.

First 50 documents

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