R. Khardon and D. Roth. Learning to reason with a restricted view. Machine Learning, 35(2):95-117, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....condition some test(s) could not be run. Thus we feel it is important for algorithms in our model to classify examples that have unspecified attributes. This requirement led us to have a three valued (versus binary) output. In the different learning to reason framework, Khardon and Roth [31] investigate the construction of a knowledge base for representing the world , i.e. some boolean function f . This knowledge base is then used to deduce (i.e. reason) if f logically implies ff where ff is a propositional query capturing the situation at hand. They consider various interpretations ....

Roni Khardon and Dan Roth. Learning to reason with a restricted view. In Proc. 8th Annu. Conf. on Comput. Learning Theory, pages 301--310. ACM Press, New York, NY, 1995.

....can be represented using only the positive features as long as the negative features all occur within the same terms. In addition, functions like 4 For example, if trigrams are the features, at most 2 are positive (active) while all the other are negatives. 5 for in depth discussion see [10]. the XOR can be represented using a disabler. Example 3 The XOR function can be represented by adding a disabler using only positive features: E = P 1 2 x 1 1 2 x 2 , E = P x 1 x 2 where the threshold = 1 for E and E. In this case, the disabler is active iff the example is (x 1 ; x ....

R. Khardon and D. Roth. Learning to reason with a restricted view. Machine Learning, 35(2):95-117, 1999.

....case in the architecture under discussion. In the instance mentioned in the previous paragraph, for example, we may choose to add the rule only if it covers a significant part of the probability space in some context of interest. A more general answer is provided by the learning to reason theory [14, 15]. There it is shown, among other things, that if the goal of learning is to be able to answer queries from a specifiable class of possibly exponentially many queries, then knowledge of a polynomial number of random examples suffices to answer queries from that class. In the current context these ....

....in this image, some properties may be derived that the system will transfer by analogy to the current instance. A generalization of this would be to retrieve several memories that share some features and determine what else they share in common. This would implement a form of learning to reason [14, 15]. In discussing the power of formal reasoning mechanisms Turing himself noted the inadequacy of reason unsupported by commonsense [37] We interpret this in the current context to mean that in a domain where the available knowledge may be inconsistent and uncertain, any long chains of ....

R. Khardon and D. Roth. Learning to reason with a restricted view. In Proc. 8th ACM Conference on Computational Learning Theory, pages 301--310, 1995.

....to the concepts of consistent and most robust extensions. The most robust extension is important in practice as it minimizes the number of corrected bits in order to have an extension. There are other possible treatments of missing bits appearing in the context of learning theory (see e.g. [6, 22, 24, 34, 36, 37, 39]) In this paper, we study the problems of deciding the existence of (and constructing) these extensions for various special classes of Boolean functions C, mainly from the view point of their computational complexity. We obtain computationally e#cient algorithms in some cases, and prove ....

R. Khardon and D. Roth, Learning to reason with a restricted view, Proceedings of COLT'95 (1995) 301-310.

....condition some test(s) could not be run. Thus we feel it is important for algorithms in our model to classify examples that have unspecified attributes. This requirement led us to have a three valued (versus binary) output. In the different learning to reason framework, Khardon and Roth [31] investigate the construction of a knowledge base for representing the world , i.e. some boolean function f . This knowledge base is then used to deduce (i.e. reason) if f logically implies ff where ff is a propositional query capturing the situation at hand. They consider various interpretations ....

Roni Khardon and Dan Roth. Learning to reason with a restricted view. In Proc. 8th Annu. Conf. on Comput. Learning Theory, pages 301--310. ACM Press, New York, NY, 1995.

....their classi cation and equivalence queries return clauses as counterexamples. We present a general transformation that allows us to obtain an entailment certi cate from an interpretation certi cate for propositional logic. Similar observations have been made before in di erent context (e.g. [18, 8]) where one transforms ecient algorithms not just certi cates. Note however, that for ef ciency we must be able to solve the implication problem for the language of hypotheses used by the algorithm. De nition 7. Let x be an interpretation. Then ones(x) is the set of variables that are set in ....

R. Khardon and D. Roth. Learning to reason with a restricted view. Machine Learning, 35(2):95-117, 1999.

....been on the Inference with Classi ers paradigm. This is a concrete instantiation of our earlier work in the Learning to Reason framework an integrated theory of learning, knowledge representation and reasoning within a uni ed framework which uses learning to facilitate high level reasoning tasks [15, 23, 16, 24, 14]. Within the Inference with Classi ers paradigm we develop inference algorithms that take as input outcomes of classi ers. Our work in this area moves away from the study of single classi ers and studies ways to use classi ers in down stream inference processes [20, 22] Speci cally, we consider ....

R. Khardon and D. Roth. Learning to reason with a restricted view. Machine Learning, 35(2):95-117, 1999.

....framework and relate it to previous work. Missing attribute values: Several other learning algorithms produce classi ers that can deal with partially speci ed instances; cf. DLR77, LR87, Qui89, SG94] However, these classi ers are not able to actively obtain missing information. Other research [BDD93, KR99, GKS97, KR95] considers the problem of learning from partially speci ed instances, but with the goal of (resp. later classifying complete instances, or later reasoning with respect to the learned concept; n.b. these other systems do not consider ways for the classi er to gather more information. Littlestone ....

R. Khardon and D. Roth. Learning to reason with a restricted view. Machine Learning, 35(2):95-117, 1999.

....framework and relate it to previous work. Missing attribute values: Several other learning algorithms produce classi ers that can deal with partially speci ed instances; cf. DLR77, LR87, Qui89, SG94] However, these classi ers are not able to actively obtain missing information. Other research [BDD93, KR99, GKS97, KR95] considers the problem of learning from partially speci ed instances, but with the goal of (resp. later classifying complete instances, or later reasoning with respect to the learned concept; n.b. these other systems do not consider ways for the classi er to gather more information. Littlestone ....

R. Khardon and D. Roth. Learning to reason with a restricted view. In Proc. 8th Annu. Conf. on Comput. Learning Theory, pages 301-310, 1995.

....to make knowledge intensive inferences would bene t if learning, knowledge representation and inference processes could be studied within a uni ed framework. This research continues an e ort to develop an integrated framework for the study of learning, knowledge representation and reasoning [10, 11, 24] and investigates a crucial stage in the computational process, studied in the context of a concrete collection of large scale problems. We are interested in learning a de nition for some relations or concepts of interest, a de nition which can be evaluated eciently given an instance in the ....

R. Khardon and D. Roth. Learning to reason with a restricted view. Machine Learning, 35(2):95{ 117, 1999.

....the learnability of strategies represented by rule based systems. In this model, the learner is given access to traces of behavior of another agent and using these traces it tries to reconstruct a strategy for behaving successfully in the same world. Following previous work on learning to reason (Khardon Roth, 1995, 1997) the formalization utilizes two general ideas. First, one can gain insights by focusing on learning that is done for the purpose of performing well in a particular task. Second, when coupling learning with the task, the competence required of the agent can be defined relative to its ....

....a third distinct value neither g i not g i are satisfied if x i = ffl The agent has at its disposal a set of actions O = fo 1 ; o n g. The symbol o i denotes the name of the action. In the learning model, the agent is not given any information 2 This is in contrast with previous work (Khardon Roth, 1995) where a special semantics is given to partial assignments. on the effects of the actions, or the preconditions for their application. In particular, there is no hidden assumption that the effects of the actions are deterministic, or that they can be exactly specified. The choice of n, the ....

Khardon, R., & Roth, D. (1995). Learning to reason with a restricted view. Proceedings of the Conference on Computational Learning Theory, pp. 301--310. Santa Cruz, CA: ACM Press.

....on dealing with incomplete information is taken. We briefly discuss two results in the learning to reason framework that use model based representations in order to exploit the relevant information in the reasoning process. In particular, it has been shown, within the Learning to Reason framework [10, 12], that modelbased representations that are suitable for the reasoning tasks considered in this paper can be learned efficiently. The model based approach to default reasoning can therefore be incorporated within an inductive setting. The model based representation can be efficiently learned, ....

R. Khardon and D. Roth. Learning to reason with a restricted view. In Workshop on Computational Learning Theory, pages 301--310, July 1995.

....can be learned efficiently from interaction with the environment, to yield efficient Learning to Reason algorithms. Later in the paper we discuss the relation of this work to the default reasoning literature. Now we briefly mention some works that are related to the approach presented here. In [Khardon and Roth, 1995b] a Learning to Reason approach that can deal with partial information is developed and shown to support efficient deduction. The interpretation taken there, however, is not expressive enough to support non monotonic reasoning. In [Khardon and Roth, 1995a] a solution to some restricted cases of ....

R. Khardon and D. Roth. Learning to reason with a restricted view. In Workshop on Computational Learning Theory, July 1995.

....attributes, as well as its classification accuracy. Missing attribute values: Several other learning algorithms produce classifiers that can deal with partially specified instances; cf. DLR77, LR87, Qui89, SG94] However, these classifiers are not able to actively obtain missing information. [BDD93, KR95] consider the problem of learning from partially specified instances, but with the goal of (resp. later classifying complete instances, or later reasoning with respect to the learned concept. GGK96] also considers learning from partially specified instances, but in situations where this missing ....

R. Khardon and D. Roth. Learning to reason with a restricted view. In COLT-95, pages 301--310, 1995.

No context found.

Khardon, R., and Roth, D. 1995b. Learning to reason with a restricted view. In Workshop on Computational Learning Theory, 301--310.

....the usefulness of this approach. These include algorithms that use model based representations to handle some fragments of Reiter s default logic as well as some cases of circuit diagnosis [20] A theory of reasoning with partial models and the learnability of such representations is studied in [21]. The question of translating between characteristic models and propositional expressions (which is relevant in database theory as well) has also been studied. Some results on the complexity of this and related questions are described in [17] Most of the work on reasoning assumes that the ....

....the usefulness of this approach. In particular, efficient algorithms for reasoning within context and for default reasoning have been developed [20] An extension of the theory presented here, that applies in the case where only partial assignments are given in the knowledge base, is described in [21]. This work is part of a more general framework which views learning as an integral part of the reasoning process. We believe that some of the difficulties in constructing an adequate computational theory to reasoning result from the fact that these two tasks are viewed as separate. The learning ....

R. Khardon and D. Roth. Learning to reason with a restricted view. In Workshop on Computational Learning Theory, pages 301--310, July 1995.

No context found.

Khardon, R. and D. Roth. 1994c. Learning to reason with a restricted view.

....represented by rule based systems. In this model, the learner is given access to traces of behavior of another agent and using these traces it tries to reconstruct a strategy for behaving successfully in the same world. Following previous work on learning to reason (Khardon and Roth, 1994; Khardon and Roth, 1995) the formalization utilizes two general ideas. First, one can gain insights by focusing on learning that is done for the purpose of performing well in a particular task. Second, when coupling learning with the task, the competence required of the agent can be defined relative to its learning ....

....agent is given some time, say N steps (where N is some fixed polynomial in the complexity parameters) to 3 Our treatment of partial assignments that include the value follows (Valiant, 1995; Roth, 1995) and simply considers the value as a third value. This is in contrast with previous work in (Khardon and Roth, 1995) where special a semantics is given to partial assignments. achieve the goal g starting with state x. In order to do this the learner has to apply its actions, one at a time, until its measurements have a value y which satisfies g (i.e. g(y) 1) Intuitively, each action that is taken changes ....

Khardon, R. and D. Roth. 1995. Learning to reason with a restricted view. In Proc. Workshop on Comput. Learning Theory, pages 301--310.

....context of the most basic reasoning task, namely deductive reasoning. The framework, however, should be seen in a more general context and can be applied for a variety of tasks. In particular, similar results have been recently developed for other, related, reasoning tasks within this framework [KR95a, Rot95, KR95b, Kha96, GGR96]. These include non monotonic reasoning, Learning to Reason with partial observations, Learning to Act in a dynamic world, and learning of active classifiers. 1.2 Comparison with Related Work The generally accepted framework for the study of reasoning in intelligent systems is the knowledgebased ....

....a result we suggest a new set of questions, oriented towards learning a functionality rather than learning a representation of the world. For the task of logical deduction we show cases where this is indeed useful. As mentioned above, similar results for other tasks have been recently obtained [KR95a, Rot95, KR95b, Kha96, GGR96]. ffl From the Knowledge Representation and Reasoning point of view the main contribution is the change of paradigm. We suggest that the reasoner interacts with the world and constructs a suitable knowledge representation. Consequently, the effectiveness of the knowledge representation depends on ....

[Article contains additional citation context not shown here]

R. Khardon and D. Roth. Learning to reason with a restricted view. In Workshop on Computational Learning Theory, pages 301--310, July 1995.

....been recently developed for 2 Notice that restricting the classes of queries considered does not change the intractability of the deduction problem, if the world is represented traditionally, as a CNF formula. Learning to Reason Delta 5 other, related, reasoning tasks within this framework [Khardon and Roth 1995a; Roth 1995; Khardon and Roth 1995b; Khardon 1996; Greiner et al. 1996] These include non monotonic reasoning, Learning to Reason with partial observations, Learning to Act in a dynamic world, and learning of active classifiers. 1.2 Comparison with Related Work The generally accepted framework ....

....that restricting the classes of queries considered does not change the intractability of the deduction problem, if the world is represented traditionally, as a CNF formula. Learning to Reason Delta 5 other, related, reasoning tasks within this framework [Khardon and Roth 1995a; Roth 1995; Khardon and Roth 1995b; Khardon 1996; Greiner et al. 1996] These include non monotonic reasoning, Learning to Reason with partial observations, Learning to Act in a dynamic world, and learning of active classifiers. 1.2 Comparison with Related Work The generally accepted framework for the study of reasoning in ....

[Article contains additional citation context not shown here]

Khardon, R. and Roth, D. 1995b. Learning to reason with a restricted view. In Workshop on Computational Learning Theory (July 1995), pp. 301--310.

....the environment. Namely, the agent can incrementally construct a representation that is relevant to its environment, in the sense that it supports correct reasoning there. This intuitive idea is formalized in a more general setting in the Learning to Reason framework [7] and is discussed also in [31, 22, 8]. We discuss two results in the Learning to Reason framework that use model based representations in order to exploit the relevant information in the reasoning process. To summarize, this paper studies reasoning with model based representations, and further substantiates the claim regarding their ....

R. Khardon and D. Roth. Learning to reason with a restricted view. In Workshop on Computational Learning Theory, pages 301--310, July 1995.

No context found.

R. Khardon and D. Roth. Learning to reason with a restricted view. In Proc. 8th ACM Conference on Computational Learning Theory, pages 301--310, 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