A. Ram and M.T. Cox. Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski and G. Tecuci, editors, Machine Learning: A Multistrategy Approach, Volume IV. Morgan Kaufman Publishers, Inc., 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....be said to have an A brain and a B brain. Roger Schank. Our work is in many ways related to that of Roger Schank and his students, especially Kris Hammond and Ashwin Ram. We drew much from their notion of failure directed learning in case based systems, as described in [SCH82] HAM90] and [RAM94]. Gerald Sussman. His 1972 Ph.D. thesis on debugging [SUS72] described perhaps the first AI system that operated by recognizing problem solving failures in order to help direct the process of recovering from the failure. The system we present in this thesis is strongly motivated by this idea, ....

Ashwin Ram and Michael Cox. Introspective reasoning using metaexplanations for multistrategy learning. In R. Michalski & G. Tecuci (Eds.), Machine learning: A multistrategy approach IV (pp. 349-377). Morgan Kaufmann: San Mateo, CA. 1994.

....ones discussed so far, which all incorporate an explicit model of reasoning separate from the main task s reasoning process. Meta AQUA performs failure driven introspective learning in a case based manner, by retrieving and applying cases which describe reasoning events instead of domain events #Ram Cox, 1994#. Meta AQUA is a story understanding system that applies explanation pattern cases #XPs# to explain anomalies #expectation failures# discovered in the story it is given. When Meta AQUA fails to understand a story fragment it expects to understand, it applies #Introspective Meta XPs #IMXPs# to ....

Ram, A. & Cox, M. #1994#. Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. & Tecuci, G. #Eds.#, Machine Learning: A Multistrategy Approach. Morgan Kaufmann.

....again, to deliver the desired path, Autognostic would have another learning opportunity, and it would repeat its blame assignment and learning task. 6 Discussion Reflective reasoning has received much attention in AI [Davis 1977, Davis 1980, Mitchell et al. 1989, Kuokka 1990, Freed et al. 1992, Ram Cox 1992] The focus of this paper, however, is the use of functional models in reflection. Hence, here we only compare our work to other research which has investigated the uses of functional models of abstract devices [Allemang 1990, Weintraub 1991, Johnson 1993] Allemang has used the Functional ....

Ram, A. and Cox, M. T., (1992) Introspective Reasoning Using Meta-Explanations for Multistrategy Learning. To appear in Machine Learning: A Multistrategy Approach IV, R. S. Michalski and G. Tecuci (eds.), Morgan Kaufmann, San Mateo, CA.

....to gathering information for argumentation (Rissland, Skalak, Friedman 1994) and on strategic methods for information retrieval (Baudin, Pell, Kedar 1994) Neither of these methods, however, learns from the search process. Our use of analogical techniques for internal reasoning is related to Ram and Cox s (1994) theory of meta explanation patterns, Kennedy s (1995) internal analogy, and Oehlmann s (1995) metacognitive adaptation. Our use of transformational analogy for case based planning, and derivational analogy for case based reasoning applied to case adaptation, combines benefits of both learning ....

Ram, A., and Cox, M. 1994. Introspective reasoning using meta-explanations for multistrategy learning.

.... Recent research on introspective reasoning has addressed the problem of representing information about reasoning methods, and applying that knowledge to the task of detecting and repairing flaws in the system s reasoning (Freed Collins, 1994; Stroulia Goel, 1994; Oehlmann et al. 1994; Ram Cox, 1994). We have developed a framework for representing reasoning methods and have implemented this approach in ROBBIE, combining an introspective reasoner with an underlying case based planner. The ROBBIE 1 system includes learning of domain knowledge through the addition of cases and learning which ....

Ram, A. & Cox, M. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. & Tecuci, G. (Eds.), Machine Learning: A multistrategy approach Vol. IV, pp. 349--377. Morgan Kaufmann.

.... knowledge and to guide acquisition of domain knowledge (e.g. Bradzil and Konolige, 1990; Davis, 1982 ] A more recent use of introspective reasoning is to monitor a system s own reasoning processes in order to refine those processes by failure driven learning (e.g. Collins et al. 1993; Ram and Cox, 1994; Cox and Freed, 1995 ] This paper presents an approach to introspective reasoning for refining the case retrieval criteria of a case based planning system. In the approach we are investigating, an introspective reasoning component monitors the processing of a case based reasoning system and ....

....its domain and reasoning tasks, and remembering the resulting successes or failures. Cox implements introspective reasoning in MetaAQUA by maintaining a set of reasoning trace templates (Meta XPs) that describe different reasoning failures, and detecting reasoning failures that match a template [ Ram and Cox, 1994 ] Meta XPs provide the means for determining the failure and suggestions for repairs. In CBR research outside of introspective reasoning, multiple methods have been proposed for determining relevant indices. For example, explanations are often used to determine the relevance of features when ....

A. Ram and M. Cox. Introspective reasoning using meta-explanations for multistrategy learning. In R. Michalski and G. Tecuci, editors, Machine Learning: A multistrategy approach Vol. IV, pages 349--377. Morgan Kaufmann, 1994.

.... as a planful process Our approach builds on the model of memory traversal and index elaboration in CYRUS (Kolodner, 1984) and especially on prior proposals for introspective failure driven learning to repair memory organization problems (Birnbaum, Collins, Brand, Freed, Krulwich, Pryor, 1991; Ram Cox, 1994). However, this model differs in treating memory search as a knowledge planning process (Hunter, 1990) In the knowledge planning framework, information search is conducted by a planning process based on explicit reasoning about needs for information and how to satisfy them. In that process, a ....

Ram, A. & Cox, M. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. & Tecuci, G. (Eds.), Machine Learning: A Multistrategy Approach. Morgan Kaufmann. In Press.

....use a model based reasoner like the one used here to introspectively improve a learning chess program, CASTLE. Stroulia Goel (1994) use a structurebehavior function model of a system s functioning to model and improve a multi strategy planning system, Router, and a design system, Kritik2. Ram Cox (1994) propose metaexplanations describing kinds of system failures as a means of doing introspective reasoning for a CBR system. Oehlmann, Edwards, Sleeman (1994) have also done work in reindexing cases and metacognitive issues for CBR. Other approaches to memory search and index learning are ....

Ram, A. & Cox, M. T. (1994). Introspective reasoning using metaexplanations for multistrategy learning. In Michalski, R. & Tecuci, G. (Eds.), Machine Learning: A Multistrategy Approach, Vol. IV. Morgan Kaufman.

....ones discussed so far, which all incorporate an explicit model of reasoning separate from the main task s reasoning process. Meta AQUA performs failure driven introspective learning in a case based manner, by retrieving and applying cases which describe reasoning events instead of domain events (Ram Cox, 1994). Meta AQUA is a story understanding system that applies explanation pattern cases (XPs) to explain anomalies (expectation failures) discovered in the story it is given. When Meta AQUA fails to understand a story fragment it expects to understand, it applies Introspective Meta XPs (IMXPs) to ....

Ram, A. & Cox, M. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. & Tecuci, G. (Eds.), Machine Learning: A Multistrategy Approach. Morgan Kaufmann.

.... 1986; Veloso, 1994, Chapter 8) This approach has attracted interest not only for domain problem solving tasks, but also in a number of systems that store and reuse reasoning traces for introspective reasoning and learning (e.g. Kennedy, 1995; Leake, Kinley, Wilson, Chapter 11; Oehlmann, 1995; Ram Cox, 1994). How to retrieve: The previous sketch used purely top down retrieval: A problem description was formed and used to select a relevant case. However, the indices needed are inextricably tied to the contents of the case library (which may change) Consequently, CBR research is also investigating the ....

.... (e.g. Kass, 1990; Kass, 1992; Schank, 1986; Schank Leake, 1986; Schank Leake, 1989) design and problem solving (e.g. Bhatta, Goel, Prabhakar, 1994; Kolodner Penberthy, 1990; Kolodner, 1994, Wills and Kolodner, Chapter 4) story generation (Turner 1994) and understanding (Moorman Ram 1994). Case based aiding systems: Case based aiding systems use automated case memories to support human reasoners. The case memories provide the experiences that human reasoners may lack, suggesting successful prior solutions and warning of prior failures. The human reasoners maintain final control, ....

[Article contains additional citation context not shown here]

Ram, A., and Cox, M. 1994. Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R., and Tecuci, G., eds., Machine Learning: A Multistrategy Approach.

.... 1986; Wilensky, 1983 ] identifying relevant features for forming useful generalizations [ DeJong, 1986; Mitchell et al. 1986 ] focusing plan repair [ Birnbaum et al. 1990; Hammond, 1989 ] guiding selection of learning methods in multistrategy learning [ Hunter, 1990b; Hunter, 1990a; Ram and Cox, 1994 ] and indexing learned information in useful ways [ Barletta and Mark, 1988; Hammond, 1989; Leake and Owens, 1986; Ram, 1993; Schank, 1982 ] Unfortunately, effectively building the explanations required by these models is a difficult problem. In order to generate any explanations at all, it ....

....of knowledge: 1. Types of goal based needs for information: In current systems that learn in response to information needs, the primary motivation for learning is the detection of processing failures or expectation failures (e.g. Birnbaum et al. 1990; Hammond, 1989; Leake, 1992; Ram, 1993; Ram and Cox, 1994; Riesbeck, 1981; Schank, 1986 ] However, information needs may arise without explicit failures. Many tasks can drive explanation, and performing those tasks may require gathering information through explanation, without being prompted by a failure per se. For example, the policy to improve ....

[Article contains additional citation context not shown here]

Ram, A. and Cox, M. 1994. Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. and Tecuci, G., editors 1994, Machine Learning: A Multistrategy Approach. Morgan Kaufmann. In Press.

....for failure driven refinement of reasoning criteria. For example, systems such as CHEF refine their indexing criteria when bad outcomes result from applying an incorrect case, and other projects apply introspective monitoring to detect reasoning failures in order to respond to them (Cox, 1994; Ram Cox, 1994). The focus of that work differs from our approach to refining retrieval, however, in that their approaches trigger learning only in response to failures in the outcomes of processing steps. Our method of index refinement enables learning to occur not only in response to bad outcomes, but and ....

Ram, A. & Cox, M. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. & Tecuci, G. (Eds.), Machine Learning: A Multistrategy Approach. Morgan Kaufmann.

....how the system reasons, and what the expected results of that reasoning are. There are several different recent approaches to the task of introspective reasoning: RAPTER (Freed Collins, 1994a, 1994b) uses expectations about a reactive planning task to diagnose and repair failures, Meta AQUA (Ram Cox, 1994) maintains a set of templates for reasoning failures with applicable repairs to apply to failed reasoning traces, Autognostic (Stroulia Goel, 1994) uses an Structure BehaviorFunction model of its own reasoning to find learning opportunities, and IULIAN (Oehlmann, Edwards, Sleeman, 1994, 1995) ....

.... Goel, 1994) applies an existing kind of model (used for modeling physical machines) to implement a self model and successfully applied the model and mechanisms to two independent systems (Kritik2 (Stroulia Goel, 1992) and Router (Goel, Callantine, Shankar, Chandrasekaran, 1991) Meta AQUA (Ram Cox, 1994) uses abstract descriptions of reasoning traces that might arise under any similar reasoning explanation task. Evaluating self modeling systems It is often problematic in AI to explain exactly what a given system has accomplished besides showing some implementation is possible. It is important ....

Ram, A. & Cox, M. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. & Tecuci, G. (Eds.), Machine Learning: A multistrategy approach Vol. IV, pp. 349--377. Morgan Kaufmann.

.... learning goals or knowledge goals of the reasoner is an important aspect of the learning problem (e.g. Collins, Birnbaum, Krulwich, Freed , 1993; desJardins, 1992; Hunter, 1990; Laird, Rosenbloom, Newell , 1986; Michalski, 1993; Mitchell, Utgoff, Banerji , 1983; Ram, 1989; Ram, 1991; Ram Cox, 1993; Ram Hunter, 1992] We argue that active, goal based learning is important for functional or computational reasons as well as for cognitive reasons. Thus formulating learning goals, asking questions, focussing attention, and pursuing learning actions are essential components of our learning ....

....learning goals, asking questions, focussing attention, and pursuing learning actions are essential components of our learning model. Since the need to learn often arises from a reasoning failure, credit or blame assignment also plays a central role in learning [Hammond, 1989; Minsky, 1985; Ram Cox, 1993; Schank, 1982; Stroulia, Shankar, Goel, Penberthy, 1992; Sussman, 1975; Weintraub, 1991] Experiential learning: Learning is an incremental process of theory formation, theory revision, and conceptual change, which occurs as the reasoner accumulates experience in some task domain. Through this ....

[Article contains additional citation context not shown here]

A. Ram and M.T. Cox. Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski and G. Tecuci, editors, Machine Learning: A Multistrategy Approach, Volume IV. Morgan Kaufman Publishers, Inc., 1993.

No context found.

Ram, A., & Cox, M. T. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In R. S. Michalski & G. Tecuci (Eds.), Machine learning IV: A multistrategy approach (pp. 349-377). San Francisco: Morgan Kaufmann.

....This paper has presented a computational theory of deliberate learning that has explored the metaphor of non linear planning as a vehicle for constructing a learning strategy. The theory represents a general approach to knowledge intensive learning, as supported by broad hand coded evaluations [75], large scale empirical experiments (reported here) and by previous psychological modelling studies [79] 25 This theory of multistrategy learning supports three major processing 62 phases that represents an intelligent system s response to reasoning failure: blame assignment, deciding what to ....

A. Ram and M. T. Cox, Introspective reasoning using meta-explanations for multistrategy learning, in: R. S. Michalski and G. Tecuci, eds., Machine Learning IV: A multistrategy approach (Morgan Kaufmann, San Francisco, 1994) 349-377.

.... by evidence from psycholinguistics (e.g. Holbrook et al. 1992; van Dijk Kintsch, 1983) reading comprehension (e.g. Black Seifert, 1981; Graesser et al. 1991) story understanding (e.g. Ram, 1991; Rumelhart, 1977) memory (e.g. Anderson, 1974; Kolodner, 1984) and metacognition (e.g. Ram Cox, 1994; Weinert, 1987) The power of this modular, integrated approach is that each supertask directly supports all of the others. No single supertask can exist in a vacuum; rather, each relies on aspects of the processing being performed by the remaining ones. Furthermore, every task in the theory is ....

....this point, the creative understandingprocess must attempt to understand the novel concept. The algorithm is a four step cycle; each pass through the algorithm increases the potential for successful understanding. An overview of this process follows; for a more complete description see Moorman Ram (1994). 1. During reading, internal representations of the text are being built and maintained. As each text phrase is decoded, the reader attempts to incorporate it into the existing structures. Memory retrieval occurs, with the new concept acting as a probe. If the concept already exists in memory, ....

Ram, A. and Cox, M. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. and Tecuci, G., editors, Machine Learning: A Multistrategy Approach, volume IV, pages 349--377. Morgan Kaufmann, San Mateo, CA.

....situations in the future correctly. This is essentially a case based or experience based approach, which relies on the assumption that it is worth learning about one s experiences since one is likely to have similar experiences in the future (see, e.g. Hammond, 1989; Kolodner Simpson, 1984; Ram, 1992; Schank, 1982] Opportunistic learning: An important corollary of the active and experiential nature of learning is that learning is opportunistic. Often, a desired piece of knowledge will not be immediately available in the input, and so the corresponding knowledge goal will not be immediately ....

....are currently active. In other words, knowledge goals can be satisfied opportunistically during the course of understanding [Birnbaum, 1986; Dehn, 1989; Hammond, 1988; Ram, 1989; Ram, 1991] leading to opportunistic learning of information previously identified as being useful to obtain (e.g. [Ram, 1992; Ram Hunter, 1992] In order for this to happen, the reasoner must be able to remember what it needs to learn, and recognize opportunities to learn the desired knowledge. Multistrategy learning: There are several things one might learn from any experience, and several different ways of ....

[Article contains additional citation context not shown here]

A. Ram & M.T. Cox. Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski & G. Tecuci (eds.), Machine Learning IV: A Multistrategy Approach, Morgan Kaufman Publishers, San Mateo, CA, 1992, to appear.

....and contrasts both computational and psychological perspectives on the phenomenon of forgetting. Section 5 closes the paper with a brief discussion. Introspective Multistrategy Learning This paper illustrates how forgetting effects learning in a multistrategy learning system called Meta AQUA (Ram Cox, 1994). The system learns by choosing a learning strategy on the basis of introspective explanations of its own performance failures. The performance task for Meta AQUA is story understanding. That is, given a stream of concepts as the representation for a story sequence, the task is to create a ....

....by choosing a proper learning strategy. An extension of explanation pattern (XP) theory (Ram, 1991, 1993; Schank, 1986) helps the system to reason about these types of failures. A meta explanation pattern (MetaXP) is an explanation of how and why an ordinary explanation fails in a reasoning system (Ram Cox, 1994). Two classes of Meta XPs exist to facilitate a system s ability to reason about itself and to assist in selecting a learning algorithm or strategy. A Trace Meta XP (TMXP) explains how a system generates an explanation about the world or itself, and an Introspective Meta XP (IMXP) explains why the ....

[Article contains additional citation context not shown here]

Ram, A., & Cox, M. T. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In R. Michalski & G. Tecuci (Eds.), Machine learning: A multistrategy approach IV (pp. 349-377). San Mateo, CA: M. Kaufmann.

....method by which it achieves these goals. These stages are detailed in Figure 1. To appear in Proceedings of the Second European Workshop on Case Based Reasoning. Chantilly, France. Nov 7 10, 1994. 2 Previous publications have dealt with the blame assignment stage (Cox, 1993; Cox Ram, 1992; Ram Cox, 1994). This paper explores how learning goals are spawned when deciding what to learn (Section 2) and how these goals are satisfied in the strategy selection phase (Section 3) A simpler system might forego explicit learning goals altogether, and directly map a failure to a learning algorithm. The ....

....reasoner at the appropriate time, as well as modify the structure or content of a concept itself. Such reorganization of knowledge affects the conditions under which a particular piece of knowledge is retrieved or the kinds of indexes associated with an item in memory. A program called Meta AQUA (Ram Cox, 1994) was written to test our theory of understanding, explanation and learning. Given the drug bust story of Figure 2, the system attempts to understand each sentence by incorporating it into its current story representation, explain any anomalous or interesting features of the story, and learn from ....

[Article contains additional citation context not shown here]

Ram, A., & Cox, M. T. (1994). Introspective reasoning using meta-explanations for multistrategy learning. In R. S. Michalski & G. Tecuci (Eds.), Machine learning: A multistrategy approach IV (pp. 349-377). San Francisco: Morgan Kaufmann.

....illustrates, a goal driven learner makes decisions about what, how, and when to learn in order to further its goals. Consequently, its learning can be considered a planful process (e.g. Hunter, 1990; Leake, to appear; Michalski Ram, to appear; Pryor Collins, 1992; Quilici, this volume; Ram Cox, 1994; Ram, Cox, Narayanan, to appear; Ram Hunter, Guiding the performance task by: ffl Determining the resources made available to the performance task ffl Guiding the control or search procedure used in the performance task ffl Guiding retrieval of plans, problem solutions, and other types of ....

.... the goal subgoal decompositional structure of the task goals, the choice of methods for achieving them and other decisions taken, the factors influencing those decisions, and descriptions of other reasoning actions (e.g. attempts to retrieve information) and their outcomes (Carbonell, 1986; Ram Cox, 1994). For example, forming an executable plan to get a good buy on a stereo requires knowing which stereo to buy and where to buy it. If the reasoner does not know, a reasoning failure occurs because current knowledge is insufficient to make a decision. At a suitable point in processing, the reasoning ....

[Article contains additional citation context not shown here]

Ram, A. & Cox, M.T. (1994). Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski & G. Tecuci, editors, Machine Learning: A Multistrategy Approach, Volume IV, Morgan Kaufman Publishers, San Mateo, CA.

....because he was interested in similar issues about Ireland. Understanding is a process of relating what one reads to the questions that one already has. These questions represent the knowledge goals of the understander, i.e. the things that the understander wants to learn [Dehn, 1989; Leake and Ram, 1993; Hunter, 1989; Ram, 1989, 1991, 1993; Ram and Hunter, 1992; Schank and Ram, 1988] The purpose of building explanations is to find answers to these questions and, thus, to arrive at a more complete understanding of the issues one is 5 interested in. However, while doing this, many new questions ....

....program. Any such rule must make a statement about the goals of the program, not just about the content of the domain. A similar argument can be made for the use of knowledge goals, or questions, to focus inference generation for understanding, explanation, or diagnosis [Ram, 1990c, 1991, 1993; Ram and Cox, 1993; Ram and Hunter, 1992; Ram and Leake, 1991] A goal based model of explanation and learning is a plausible account of human behavior, and also has computational advantages for the design of learning programs. 6 Our theory of questions is based on a theory of understanding tasks. In addition to ....

[Article contains additional citation context not shown here]

A. Ram and M.T. Cox. Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski and G. Tecuci, editors, Machine Learning IV: A Multistrategy Approach. Morgan Kaufman Publishers, 1993, in press.

.... an explicit model of what it needs to know to complete its understanding of the problem, i.e. of the gaps in its knowledge base, c) learn by filling in these gaps when the information it needs becomes available, and hence (d) gradually evolve a better understanding of the domain [Ram, 1990b; Ram, 1992]. Thus the learning process is focussed by the knowledge goals of the system. Reading can be thought of as one type of knowledge action. More sophisticated planners might manage a complex and interacting set of learning goals and available knowledge actions, making decisions about when to pursue a ....

....decisions about when to pursue a particular goal, based on its relationship to the program s other learning and performance goals and on on the current state of the world. These issues are being explored further in the INVESTIGATOR [Hunter, 1990b; Hunter, 1990a] and META AQUA [Cox and Ram, 1991; Ram and Cox, 1992] projects. 7 Comparison to other approaches Other cognitive theories have also included reference to desires for knowledge, although there are significant differences between those prior theories and our theory of knowledge goals. The conceptual dependency representation proposed by Schank and ....

A. Ram and M. Cox. Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R. Michalski and G. Tecuci, editors, Machine Learning IV: A Multistrategy Approach. Morgan Kaufman Publishers, Inc., 1992. In preparation.

....metareasoning task will enable the reader to answer. Metareasoning results from keeping explicit representations of the reasoning processes occurring during the cognitive activity. These reasoning traces can then be reasoned about, much like any other piece of knowledge within a cognitive agent (Ram Cox, 1993). This functionality enables a reader to reflect on their own reasoning, thus enabling them to learn more from the experience. For example, if the reader discovers that they were led down the garden path at some point in the story, the metaanalysis will enable the reader to learn this and ....

....use the creative understanding perspective to evaluate past systems. While most systems restricted themselves to strict reliance on world knowledge, there were some exceptions. AQUA built explanations of scenarios which it did not possess knowledge of, such as terrorist suicide bombings. Meta AQUA (Ram Cox, 1993) attempts to modify its world knowledge in pursuit of the interpretation which holds more of the story coherently together. Looking over past systems, one can see that the ones which performed some level of creative understanding were presented as learning systems, for the most part. A person, it ....

Ram, A. and Cox, M. (1993). Introspective reasoning using meta-explanations for multistrategy learning. In Michalski, R. and Tecuci, G., editors, Machine Learning: A Multistrategy Approach, volume IV. Morgan Kaufman Publishers, San Mateo, CA. In Press.

....based on two key ideas. First, we view learning as an active process involving the formulation of learning goals during the performance of a reasoning task, the prioritization of learning goals, and the pursuit of learning goals using multiple learning strategies (Hunter, 1990; Ram, 1989, 1991; Ram Cox, 1994; Ram Hunter, 1992) However, while previous efforts have focused on the process of generating learning goals (Cox Ram, 1994; Ram, 1991; Ram Cox, 1994) and on the planful pursuit of learning goals (Cox Ram, 1994; Hunter, 1990; Ram Hunter, 1992; Redmond, 1992) little attention has been ....

.... of a reasoning task, the prioritization of learning goals, and the pursuit of learning goals using multiple learning strategies (Hunter, 1990; Ram, 1989, 1991; Ram Cox, 1994; Ram Hunter, 1992) However, while previous efforts have focused on the process of generating learning goals (Cox Ram, 1994; Ram, 1991; Ram Cox, 1994) and on the planful pursuit of learning goals (Cox Ram, 1994; Hunter, 1990; Ram Hunter, 1992; Redmond, 1992) little attention has been paid to the fundamental learning actions that actually carry out the inferences necessary to learn. In order to develop a general ....

[Article contains additional citation context not shown here]

Ram, A. & Cox, M.T. (1994). Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski & G. Tecuci, editors, Machine Learning: A Multistrategy Approach, Volume IV, Morgan Kaufman Publishers, San Mateo, CA.

....to adapt selected behaviors to the immediate demands of the environment. The second case study is based on a computer program called ISAAC (Integrated Story Analysis And Creativity) which is a natural language understanding system that reads short stories from the science fiction genre (Moorman Ram, 1994a, 1994b) Such stories require creative understanding, in which the reader must learn enough about an alien world in a short text in order to accept it as the background for the story, and simultaneously must understand the story itself. ISAAC implements a process of extrapolative conceptual change ....

.... Golding, Long, 1991) story understanding (e.g. Birnbaum, 1986; Ram, 1991; Rumelhart, 1977) episodic memory (e.g. Kolodner, 1984; Schank, 1982) analogy (e.g. Falkenhainer, 1987; Gentner, 1989) creativity (e.g. Gruber, 1989; Schank Leake, 1990) and metacognition (e.g. Gavelek Ram Cox, 1994; Raphael, 1985; Schneider, 1985; Weinert, 1987; Wellman, 1985) The supertasks and their functions are summarized below; a high level system architecture is shown in Figure 5. Story Text Scenario Comprehension Comprehension Story Structure Metacontrol Sentence Processing Memory Management ....

[Article contains additional citation context not shown here]

Ram, A. & Cox, M.T. (1994). Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski & G. Tecuci, editors, Machine Learning: A Multistrategy Approach, Volume IV, Morgan Kaufman Publishers, San Mateo, CA.

....during planning or comprehension, then it must take the problem of forgetting seriously if it is to learn. A failure driven learning system gains experience by adjusting its background knowledge (BK) in response to errors, so as to avoid repeating similar failures in the future. As argued in Ram and Cox (1992), the organization of the BK, as well as the BK itself, are possible causes of failures during the reasoning process. Blame assignment involves determining the cause of failure in order to decide what to learn. Thus an analysis of forgetting is essential for effective blame assignment, and for ....

....and G. E. Lasker (eds. Proceedings of the Sixth Inernational Conference on Systems Research, Informatics and Cybernetics, Baden Baden, Germany (August, 1992) pp 115 120. 116 The solution for this type of learning is to represent the reasoning process explicitly in structures called Meta XPs (Ram and Cox, 1992). A Trace Meta XP (TMXP) is a structure that records a reasoning trace and explains how solutions were generated, whereas an Introspective Meta XP (IMXP) is a causal pattern that, when applied to a TMXP, explains why these solutions fail. These structures allow direct inspection of the reasoning ....

[Article contains additional citation context not shown here]

Ram, A., and Cox, M. T. (1992); Introspective Reasoning Using Meta-Explanations for Multistrategy Learning.

....a system to improve its own reasoning as well as its domain knowledge. Our model is based on the proposal of (Birnbaum et al. 1991) to use a model of the ideal behavior of a case based system to judge system performance and to refine its reasoning mechanisms; it also draws on the research of (Ram Cox 1994) on introspective failure driven learning. This work examines introspection guided by expectation failures about reasoning performance. We are developing a vocabulary of failures for the case based system, an introspective reasoner which uses a hierarchical model of system behavior, and a method ....

....of failure types to guide our choice of assertions to include in the model. For example, identifying the failure failing to complete adaptation leads to assertions about how to gauge the progress of adaptation in this planner. We also include higher level failure types as are described in (Ram Cox 1994); some such failures recurred for different components of the planner, leading us to use CBR to implement components themselves. We have constructed a skeletal hierarchical model and have begun testing the case based planner with and without introspective corrections. Initial experimental results ....

Ram, A., and Cox, M. T. 1994. Introspective reasoning using meta-explanations for multistrategy learning.

....of when and what to learn. In this way, goal driven learning follows the spirit of research on failure driven learning systems, in which learning is motivated by deficiencies in system performance (e.g. Sussman, 1975; Riesbeck, 1981; Schank, 1982; Collins Birnbaum, 1988; Hammond, 1989; Ram Cox, 1993; Schank Leake, 1989) Likewise, it is in the spirit of explanation based learning research on how to form useful target concepts (Kedar Cabelli, 1987) and on judging the utility of learning (e.g. Keller, 1987; Minton, 1988) Goal driven learning, however, takes a broader view, examining the ....

.... 1991) and in AI, a growing body of recent research presents functional justifications for making decisions about the usefulness of potential learning and for guiding learning according to learner goals (e.g. desJardins, 1992; Hunter, 1990; Krulwich, Birnbaum, Collins, 1992; Leake, 1992; Ram Cox, 1993; Ram Hunter, 1992; Ram, 1991) At the Fourteenth Annual Conference of the Cognitive Science Society, a symposium was organized to bring together researchers addressing goaldriven learning from diverse perspectives. The symposium provided a forum to present recent results and new directions in ....

[Article contains additional citation context not shown here]

Ram, A. & Cox, M., 1993. Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In Machine Learning: A Multistrategy Approach, eds. R. Michalski and G. Tecuci. San Mateo, CA: Morgan Kaufmann. Forthcoming.

.... goals of the reasoner is an important aspect of the learning problem (e.g. Collins, Birnbaum, Krulwich, and Freed , 1993; desJardins, 1992; Hunter, 1990; Laird, Rosenbloom, and Newell , 1986; Michalski, 1993; Mitchell, Utgoff, and Banerji , 1983; Quilici, to appear; Ram, 1989; Ram, 1991; Ram and Cox, 1994; Ram and Hunter, 1992; Redmond, 1992] We argue that active, goal based learning is important for functional or computational reasons as well as for cognitive reasons. Thus formulating learning goals, asking questions, focussing attention, and pursuing learning actions are essential components ....

....learning goals, asking questions, focussing attention, and pursuing learning actions are essential components of our learning model. Since the need to learn often arises from a reasoning failure, credit or blame assignment also plays a central role in learning [Hammond, 1989; Minsky, 1985; Ram and Cox, 1994; Schank, 1982; Stroulia, Shankar, Goel, and Penberthy , 1992; Sussman, 1975; Weintraub, 1991] Experiential learning: Learning is an incremental process of theory formation, theory revision, and conceptual change, which occurs as the reasoner accumulates experience in some task domain. Through ....

[Article contains additional citation context not shown here]

A. Ram and M. T. Cox. Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R. S. Michalski and G. Tecuci, editors, Machine Learning: A Multistrategy Approach, Volume IV. Morgan Kaufman Publishers, Inc., 1994.

....attempted ffl Aiding evaluation of results of learning with respect to the desired output Guiding storage, by: ffl Selecting what to store ffl Determining how learned knowledge is indexed Table 1: Ways in which goals can influence learning. Ram, chapter 21; Pryor Collins, 1992 chapter 10; Ram Cox, 1994 chapter 7; Ram, Cox, Narayanan, chapter 18; Ram Hunter, 1992 chapter 4; Redmond, 1992; Quilici, in press; Schank Abelson, 1977; Xia and Yeung, 1988 chapter 12) This learning process is analogous to models of problem solving in which the reasoner uses task goals to formulate action plans ....

....3; Tambe, Newell, Rosenbloom, 1990) provide support for this argument. Active, goaldriven learning implies the ability to make explicit decisions about what, when, and how to learn (Ram, Cox, Narayanan chapter 18) Thus some of the motivations for goal based approaches include (see also Cox Ram, 1994): ffl Alleviating problems of computational complexity: The ability of a reasoner to make decisions about its reasoning and learning processes helps to alleviate problems caused by the computational complexity of reasoning in an open world, by enabling the reasoner to focus its efforts towards ....

[Article contains additional citation context not shown here]

Ram, A. & Cox, M.T. (1994/chapter 7 of this volume). Introspective Reasoning using Meta-Explanations for Multistrategy Learning. In R.S. Michalski & G. Tecuci, editors, Machine Learning: A Multistrategy Approach, Volume IV, Morgan Kaufman Publishers, San Mateo, CA.

No context found.

Ram, A. and Cox, M. T. 1992. Introspective Reasoning Using Meta-Explanations for Multistrategy Learning. To appear in Machine Learning: A Multistrategy Approach IV, R. S. Michalski and G. Tecuci (eds.), Morgan Kaufmann, San Mateo, CA.

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