R. S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111--151, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....not to generalize over situations. For this reason, most work on abduction involves reasoning from observed facts to (all ground) abductive hypotheses. There is, however, some work on expanding the notion of abduction in order to support reasoning from general observations to general hypotheses [Michalski, 1993] . For example, the general observation all the students here are wearing pink corresponds to a universally quantified description. It might provide the basis for reasoning toward a general rule rather than a typical abductive hypothesis. In such cases, it is a rule rather than a fact that is ....

....all the students here are wearing pink corresponds to a universally quantified description. It might provide the basis for reasoning toward a general rule rather than a typical abductive hypothesis. In such cases, it is a rule rather than a fact that is abductively inferred, and the claim [Michalski, 1993] is that this combines inductive as well as abductive reasoning. In our question answering framework, it is possible to express generalizations in the hypothetical component of an answer: Skolem constants are taken to be universally quantified in the hypothetical component, and there is nothing ....

[Article contains additional citation context not shown here]

R. S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111--151, 1993.

....from cluster examples will contribute to the progress of research in any field involving the mapping of raw sensor signals to abstract notions of objects. We have discussed a number of example domains already, and the technique may also be applicable to problems such as multistrategy learning [16], the data mining [1] and the identification of genes in DNAs [4] The rest of this paper will therefore rise to this challenge by presenting our algorithm for LCE. Formalization of Learning from Cluster Examples This chapter formally states the task of learning from cluster examples. This task ....

R. S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111--151, 1993.

....from cluster examples will contribute to the progress of research in any field involving the mapping of raw sensor signals to abstract notions of objects. We have discussed a number of example domains already, and the technique may also be applicable to problems such as multistrategy learning [16], the data mining [1] and the identification of genes in DNAs [4] The rest of this paper will therefore rise to this challenge by presenting our algorithm for LCE. Formalization of Learning from Cluster Examples This chapter formally states the task of learning from cluster examples. This task ....

R. S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111--151, 1993.

.... set of data can be used to compress the data the more regularities there are in the data, the more them can be compressed (see references listed in [Gr u98] and [VL97, LV97] for overviews) reinforcement learning and learning in AI (two incomplete lists are [CB97, Bou96, BGS 91, DE97, Mic93] and [SSH94, Sia91, Wei93, Wei95] Learning, rationality and knowledge As far as we know, a zero one paradigm of coordination was rst introduced in formal learning theory by [MO99b] by using the tools of recursion theory. Kel96] advanced a similar paradigm (see the Frank and Gertrude game) ....

R. S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111-151, 1993.

....regarding this definition and previous work. First, hypothetical answers, and, in general, abductive hypotheses, have been identified with all ground rules. Some work on expanding the notion of abduction in order to support reasoning from general observations to general hypotheses has been done [Michalski, 1993] . The proposed definition imposes no limitations on the form of clauses other than the variable sharing property defined above. Second, the definitions provided for generic and hypothetical answers completely clarify the way in which clauses corresponding to rules should be classified: rule and ....

R. S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111--151, 1993.

....(KDD) it is very often the pre processing stage, namely transformation and reduction of data, that play an important role. A good transformation reduction technique may result in new attributes that are more appropriate for a data mining algorithm. A wide range of constructive induction techniques [6, 7, 9, 15] has been explored aiming at enriching the language of propositional learners [4, 5] The usual way is to add new attributes that are computed from existing ones using a priori given functions. One important class of functions includes linear combinations of numerical attributes. To prevent a ....

Michalski, R.S.: Inferential Theory of Learning as a Conceptual Basis for multistrategy Learning. Machine Learning(Special Issue on Multistrategy Learning) 11, 1993

....of AI class are two different reasonings because of (I2) and (A2) but related by the AI explanatory reasoning. 3 DISCUSSION AND CONCLUSION Is induction a form of abduction Is abduction a form of induction or are they different and how are they related Following Michalski s definitions [12] abduction is a form of induction (with the inductive inference is defined as hypothezing premise P, given BK and C such as P BK j= C ) However, if we consider Pierce s formulation [15] of abduction: The surprising fact, C, is observed, but if A were true, C would be a matter of course, hence ....

R. Michalski, 1993. Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning, in Machine Learning 11, pp11-151

.... of completeness the problem domain, is updated to a new theory T 0 in the light of new observations O so that T 0 captures O (i.e. T 0 j= O) At the simplest level, abduction and induction simply co exist and both function as revision mechanisms that can be used in developing the new theory [10]. In a slightly more cooperative setting, induction provides new foreground knowledge in T for later use by abduction. At a deeper level of cooperation, abduction and induction can be integrated together within the process of constructing T . There are several ways in which this can happen within ....

....when, first, through the use of basic abduction, the original observations are transformed to data on abducible background predicates in T , becoming training data for induction on these predicates. We can illustrate this type of integration with the following simple example originating from [10]. We have the observation that: O: all bananas in this shop are yellow, and the given theory T containing the statement: T : all bananas from Barbados are yellow. 5 T T O O TH = O Abduction Induction Figure 1: The cycle of abductive and inductive knowledge development. The theory T is ....

Ryszard S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111--151, 1993.

.... and Induction: an AI perspective Yannis Dimopoulos and Antonis Kakas 1 1 Introduction In this short note 2 , we compare the two separate forms of reasoning, induction and abduction, aiming at their effective integration within Artificial Intelligence (e.g. 1] 2] 6] 8] [5]) The two inferences, abduction andinduction, as they are commonly used in Artificial Intelligence, share the same basic formal specification: In a logic based approach, given a background knowledge T (often called domain theory) that describes our problem domain, and an observation (or a set ....

....generated (extracted) directly from the given theory T based only on the general laws that can be found in T . This means that abductive explanations typically do not contain new general rules. In contrast to this restricted view of abduction the term abduction has been used in some cases (e.g. [5]) more liberally allowing the inclusion of new general laws in the explanation. We argue here that in these cases the reasoning involved has a complex form that can be broken down to a combination of more basic abduction steps, of the restricted nature described above, together with inductive ....

[Article contains additional citation context not shown here]

R. S. Michalski. Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning. Machine Learning, Vol. 11, pp 111-151, 1993.

....Recent years have witnessed a growing interest in developping multistrategy learning systems that integrate two or more inferences types and or computational paradigms (ML, 1993) Such systems take advantage of the complementarity of different inference types of representational mechanisms. Michalski (1993) describes different types of inference encounter in machine learning (ML) systems. Among those inference types, there are induction and abduction which are, with deduction, the basic logical inferences. While deduction is tracing forward implications, abduction and induction are tracing backward ....

.... reversed deduction. Induction is used to synthetize information from a set of observations into a general hypothesis. Abduction allows to assume a fact if it is a possible explanation for another fact to be explained. Many works have been done in order to define both inferences (Pierce, 1898; Michalski, 1993; Ganascia, 1994; Torasso al. 1995) and their usability in different domains where hypothetical reasoning is important (AI, 1990; Barboux, 1990; Codognet, 1994; Satoh, 1994; Eiter, 1995) This paper focusses on a somehow different point which consists in finding relationships between abduction ....

[Article contains additional citation context not shown here]

Michalski, R. S. 1993. Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning. Machine Learning 11(2-3): 111-151.

....version spaces can be combined. It is also necessary to show how deductive inferences can be combined with inductive inferences. Hirsh (1989) proved several relevant results for version spaces manipulation. We describe these and other related work in the next section. We also borrow heavily from Michalski s (1993) Inferential Theory of Learning (ITL) Michalski proposed a set of transformation operators that employ deductive and inductive inference to transform existing knowledge into new knowledge. We have adapted a number of these operators to be used for manipulating and communicating version spaces. ....

....of a learning task may be distributed, including the spaces and search orders. However, as we are interested in the communication of inferences, not the distribution of the inference process, then it is unnecessary to address this in this proposal. 2.3. The Inferential Theory of Learning Michalski s (1993) Inferential Theory of Learning describes a unified theory of learning. Simply put, the theory states there are two types of inference: deductive and inductive, which give rise to other forms of reasoning such as analogy and abduction. These occur when measures of contingency are introduced. ....

R. S. Michalski (1993). Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning, Machine Learning, 11:111-151.

....I. J. Hampton, R.S. Michalski and P. Theuns, editors, 67 95, Academic Press. Hanson, S.J. and M. Bauer, 1990) Conceptual clustering, categorization and polymorphy. 3(4) 343 372. Kodrato , Y. V.S. Moustakis, and N. Graner. 1994) Can Machine Learning Solve my Problem , 8(1) 1 36. Michalski R.S. 1974) A logic based approach to optimal classi cation into a large number of classes. ECAI 92 Proceedings Logic Minimization Algorithms for VLSI Synthesis Classi cation and regression trees Proceedings of ECML 91 Proceedings of ECML 91 Machine Learning Fundamentals of Digital Logic ....

Michalski R.S., (1993b), Inferential Theory of Learning as Conceptual Basis for Multistrategy Learning, , 11(2/3), 111-152.

....which is appropriate from a knowledge engineering perspective and which also allows efficient hierarchical problem solving based on dropping sentences. We take these observations as the motivation to develop a more general model of abstraction in problem solving. As already pointed out by Michalski (1994), abstraction, in general, can be seen as switching to a completely new representation language in which the level of detail is reduced. In problem solving, such a new abstract representation language must consist of completely new sentences and operators and not only of a subset of the sentences ....

....et al. 1995) support these observations very well. In general, abstraction is the task of transforming a problem or a solution from a concrete representation into a different abstract representation, while reducing the level of detail (Michalski Kodratoff, 1990; Giunchiglia Walsh, 1992; Michalski, 1994). However, in most hierarchical problem solvers, the much more limited view of abstraction by dropping sentences is shown to be the reason why efficient ways of abstracting a problem and a solution are impossible (e.g. see Section 2.1 and Figure 4) The second weakness of most hierarchical ....

[Article contains additional citation context not shown here]

Michalski, R. S. (1994). Inferential theory of learning as a conceptual basis for multistrategy learning. In Michalski, R., & Tecuci, G. (Eds.), Machine Learning: A Multistrategy Approach, No. 11, chap. 1, pp. 3--62. Morgan Kaufmann.

....generalization which it combines to form learning plans. We argue, however, that a system capable of a truly broad and flexible range of learning needs to be able to construct its own learning algorithms from more basic components. One approach to such basic components can be found in [5] which describes a taxonomy of knowledge transmutations which can be thought of as basic operations over knowledge elements. An example of one of these transmutations is generalize. The generalize transmutation specifies that given a fact about a knowledge element one can infer that information ....

R. Michalski, Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning. Machine Learning, 11, 1993.

....terms across ontologies. The OBSERVER system [MKSI96] using synonym relationships is a step in this direction. A proposal to extend the system by using hyponym and hypernym relationships has been presented in [MKIS96] We plan to extend the system to utilize knowledge transmutation operators [Mic93] to express correspondences between terms in the various ontologies in the future. ....

Michalski R. (1993) Inferential theory of learning as a conceptual basis for multistrategy learning.<F3.17e+05> Machine Learning<F3.733e+05> 11

....the current one was unable to express the target (i.e. all concepts were inconsistent with the examples) However, it was limited to only weakening its bias, which will eventually degrade overall performance, without a corresponding strengthening operation. Multistrategy or task adaptive learners [29] choose from a fixed set of biases in a problem dependent manner. For example, 38] used a chain of learners, from explanation based (strongly biased) to theory driven (less biased) to similarity based (weakly biased) which were applied in turn until one produced a consistent concept. Problems ....

R. S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11:111--151, 1993.

....Frameworks to Predicate Invention : 16 6 Criticism against Constructive Induction and Continuing Learning 16 7 Autonomy vs. Human Intervention in Predicate Invention 17 8 Discussion 20 9 Conclusion and Further Research 21 1 Introduction [Michalski 93] distinguishes between selective induction and constructive induction as follows: Empirical learning uses little domain knowledge, while constructive induction uses more domain knowledge. A more precise way to characterize this distinction is that in empirical induction the description space for ....

Michalski R.S.: Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning, in Special Issue on Multistrategy Learning, Machine Learning, 11(2/3), 1993.

....concept hierarchies. The two type constructor attributes allow concept formation to take place in many domains [3] In particular, several data models use the same mechanisms to represent complex data [2] Another aspect of structured concepts is their relation to abstraction, as defined in [4]. In this framework, abstraction is a knowledge transmutation that decreases the level of detail at which objects are observed. This was exactly the motivation for structured objects, i.e. to capture the different levels of detail. Knowledge transmutations like concretion would be, in our ....

Michalski, R.S. (1993). Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning. Machine Learning, 11, pp 111--151.

.... represented as methods for achieving learning goals, can be chained, composed, and optimized, resulting in learning plans that are created dynamically and pursued in a flexible manner (Cox, 1993; Cox Ram, this volume; Gratch, DeJong, Chien, this volume; Hadzikadic Yun, 1988; Hunter, 1990; Michalski, 1993; Michalski Ram, to appear; Ram Hunter, 1992; Redmond, 1992; Stroulia Goel, this volume) ffl Increasing the flexibility of learning: In situations involving multiple reasoning failures, multiple active and suspended learning goals, multiple applicable learning strategies, and limited ....

....In the knowledge planning process, explicit reasoning is done about learning goals, their relative priorities, and strategies by which they can be achieved. These learning goals, also called knowledge goals (Ram, 1987, 1990; Ram Hunter, 1992) can be represented in a goal dependency network (Michalski, 1993; Michalski Ram, to appear) which is used to select and combine learning actions into learning strategies that are appropriate for current learning goals and for the learning opportunities provided by the current environment. Individual learning actions may include performing knowledge ....

[Article contains additional citation context not shown here]

Michalski, R.S. (1993). Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning. Machine Learning, 11(2/3):111--151, 1993.

....is appropriate from a knowledge engineering perspective and which also allows efficient hierarchical problem solving based on dropping sentences. We take these observations as the motivation to develop a more general model of abstraction in problem solving. As already pointed out by Michalski [ Michalski, 1994 ] abstraction, in general, can be seen as switching to a completely new representation language in which the level of detail is reduced. In problem solving, such a new abstract representation language must consist of completely new sentences and operators and not only of a subset of the ....

R. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. In R. Michalski and G. Tecuci, editors, Machine Learning: A Multistrategy Approach, number 11, chapter 1, pages 3--62. Morgan Kaufmann, 1994.

....methodsto utilize the backgroundknowledge. Mostof this work is concernedwith the problem of how to integrate abductive and inductive reasoning methods to enhance the learning capabilities of their frameworks. For a general discussion of this issue of integrating abduction and learning see [1] [6] and [4] and references therein. To ourknowledgethe only case where abductioninmachine learning has been usedas in this paper, as an integral partof the defi nition of the learning problem, is [12] The authors argue for the usefulness of abduction for learning to diagnose in situations where ....

R. S. Michalski. Inferential Theory of Learning as a Conceptual Basis for MultistrategyLearning. MachineLearning,11:111-151, 1993.

....generated (extracted) directly from the given theory T based only on the general laws that can be found in T . This means that abductive explanations typically do not contain new general rules. In contrast to this restricted view of abduction the term abduction has been used in some cases (e.g. [31]) more liberally allowing the inclusion of new general laws in the explanation. We argue here that in these cases the reasoning involved has a complex form that can be broken down to a combination of more basic abduction steps, of the restricted nature described above, together with inductive ....

....the boundary conditions for each object (e.g. the initial position and velocity) in the world. not sweet we can explain the observation(s) O = all bananas in the shop are not sweet by the abductive explanation that all bananas in the shop are from Barbados ( this example is taken from [31]) But we can also see this instead of a single process of abduction as comprising of two processes as follows. First we have a simple form of abduction to explain a typical case of a banana being not sweet and then we use (again a simple form of) induction to synthesize these different ....

R. S. Michalski. Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning. Machine Learning, Vol. 11, pp 111-151, 1993.

....The second approach is more favorable because of flexibility it provides and because it gives us additional information for relevance judgment. It is this additional information which allows us to answer the question how are the items similar . Similarity theories developed so far (e.g. [Tve77, Hol85, Lea92, SOS92, Mic93]) do not support relevance assessment if the context is changed. In order to support flexible retrieval of relevant information, we include context in our definition of similarity. 2 The Role of Context Similarity judgments are always made with respect to representations of entities, not with ....

....retrieval of relevant information, we include context in our definition of similarity. 2 The Role of Context Similarity judgments are always made with respect to representations of entities, not with respect to the entities themselves [MO89] It is known that similarity is context dependent [Mic93]. Systems with predefined similarity relations predefine context as well. However, if the formalism for similarity assessment does not capture context (e.g. Tve77] then one cannot model similarity changes flexibly. Context in similarity allows for finding information approximation by attention ....

Ryszard S. Michalski. Inferential theory of learning as a conceptual basis for multistrategy learning. Machine Learning, 11(2):3--151, 1993.

....that is, to develop a principled theory of the learning actions involved in acquiring and transforming knowledge. The second key idea, then, is to model learning as a kind of inference in which the system augments and reformulates its knowledge using various types of primitive inferential actions (Michalski, 1993b) These primitive inferences, known as knowledge transmutations, include generalization, abstraction, explanation, and similization, and their counterparts, specialization, concretion, prediction, and dissimilization. For example, the familiar kind of learning known as concept formation is ....

....Knowledge transmutations can be combined in a flexible and dynamic manner to yield the desired learning behavior as specified by the system s learning goals. This view of learning is known as the inferential theory of learning, since it views learning fundamentally as a process of inference (Michalski, 1993b) In contrast to broad grained characterizations of learning in terms of traditional machine learning and information gathering algorithms (such as explanation based generalization, inductive generalization, or database lookup, typically represented as modules in a multistrategy learning ....

[Article contains additional citation context not shown here]

Michalski, R.S. (1993b). Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning. Machine Learning, 11(2/3):111--151, 1993.

No context found.

R. S. Michalski (1993). Inferential Theory of Learning as a Conceptual Basis for Multistrategy Learning, Machine Learning, 11:111-151.

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