Veloso, M. M. 1992. Learning by Analogical Reasoning in General Problem Solving. Ph.D. Dissertation, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA. Available as technical report CMU-CS-92-174. A revised version of this manuscript will be published by Springer Verlag, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... in which rules and cases are dependent, meaning that one representation scheme was derived from the other (i.e. rules derived from cases or vice versa) and the efficiency of the integrated scheme exceeds the efficiency that could have been achieved with rules or cases alone (e.g. 3] 15] [22]) The latter involves approaches in which the two representation schemes are independent and their integration results in improved accuracy compared to each one representation scheme working individually (e.g. 5] 11] 20] A popular integration method resulting in accuracy improvement is ....

Veloso, M.M.: Learning by Analogical Reasoning in General Problem Solving. PhD Thesis, Technical Report CMU-CS-92-174, Carnegie Mellon University, Pittsburg, PA (1992).

.... Selective utilization filters [Markovitch Scott, 1989] prevent the problem solver from seeing certain knowledge items for some period of time, and are often implemented in CBR systems as indexing schemes [Kolodner, 1994] or by time bounding the case retrieval mechanism [Brown, 1993; Veloso, 1992] In these approaches only a subset of the case base is made available to the retrieval mechanism with the attendant disadvantage that certain cases may be missed at retrieval time. Unfortunately, these omissions can result in unnecessarily complex adaptation stages or even problemsolving ....

M. Veloso. Learning by Analogical Reasoning in general Problem Solving. Ph.D Thesis (CMU-CS-92-174). Carnegie Mellon University, Pittsburgh, USA, 1992.

.... domain independent (shown in Table 3) and domain dependent (shown in Table 4) This second grammar is generated on the y by EvoCK for any domain described in Prodigy4.0 Description Language [6] The grammar shown in Table 4 is for the logistics domain, a well known planning domain described in [40]. This domain represents a setup where packages have to be delivered to several locations using trucks (inside a city) or airplanes (between cities) Terminal symbols in the grammars are displayed in lowercase and non terminal generative symbols are shown in uppercase. This grammar has production ....

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, School of Computer Science (Carnegie Mellon University), 1992.

.... using C , a simple best first planner that uses the GRT heuristic to guide the search process (we call it GRT planner henceforth) We ran this planner on several classical problem instances, taken from the bibliography, such as some blocks world instances [8] some logistics instances [13] and some rocket instances [2] Moreover, we ran GRT planner on some new problem instances originally presented at the AIPS 98 planning competition. Our planner was faster in all of the cases, finding also in most (but not all) of the cases shorter solutions. The rest of the paper is organized ....

....each ground fact p an integer i that estimates the steps needed to achieve p from the goal state, as it is described in the next sub section. A problem when calculating H GRT backwards is that the goal state in most of the problems is incomplete. For example, in the logistics domains [13] the goal state does not determine where are the trucks and the planes. If the goal state is incomplete, it is impossible to apply the inverted operators to it and hence to estimate the distances of the domain s facts from the goal. The solution adopted is to enrich the incomplete goal state with ....

[Article contains additional citation context not shown here]

Veloso M.: Learning by Analogical Reasoning in General Problem Solving. PhD Dissertation, Computer Science Dept., CMU Tech. Report, CMU-CS-92-174 (1992)

....it is briefly presented the application of the MO GRT planner in a transportation logistics domain. A more detailed presentation can be found in [17] and [23] The logistics MO domain In order to demonstrate the efficiency of the MO GRT planner, it has been used as a basis the logistics domain [24], which is commonly used in the bibliography and in the planning competitions (AIPS 98 and AIPS 00) In this domain there are several cities, each one containing several locations, some of which are airports. There are also trucks, which can move within a single city, and airplanes, which can fly ....

Veloso M. Learning by analogical reasoning in general problem solving. Ph.D. thesis

....e ect. Furthermore, the above constraint can exist independently of any negative examples. As a counterpart, the diculty is to nd the relevant constraints that would compensate correctly for the rules of the theory [3] This is part of our current work. Let us consider the Logistics domain [15]. In this domain we have two types of vehicles: trucks and airplanes. Trucks can be used to transport goods within a city, and airplanes can be used to transport goods between two airports. The problems in this domain typically start o with a collection of objects at various locations in various ....

M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Department of Computer Science, Carnegie Mellon University, 1992.

....efficiently around 10 6 states) in less than 10 seconds is a very encouraging result. Even if comparable with graphplan on the fixit problem, mbp seems in general to decrease performance in domains which can be solved with plans of parallel actions. We have performed a test on logicistic [ Veloso, 1992 ] a domain where most of the actions can be performed in parallel. logicistic.a and logicistic.b are two examples with different numbers of cities and packages (5 and 4, and 3 and 6, respectively) These examples are rather hard. ucpop was unable to find a solution even for problems with 3 ....

Manuela Veloso. Learning by analogical reasoning in general problem solving. PhD thesis, 1992. CMU, CS Techn. Report CMU-CS-92-174.

....weights and their scales play in the planning process, i.e. how they affect the planning time and the quality of the resulting plans. This is performed through an adequate number of experiments in an enhanced logistics type domain. 5. 1 The logistics MO Domain In the original logistics domain [19] there is a single means of transportation to transfer an object between two cities: the airplane. In order to measure the effectiveness of MO GRT, we have extended this description with trains, which can only perform transportations between different cities, and we have characterized one location ....

M. Veloso, Learning by analogical reasoning in general problem solving, Ph.D. Thesis, Department of Computer Science, Carnegie Mellon University, 1992.

....two notions of length may actually be in conflict, where the parallel minimum solution has non minimum sequential length. 12 We will investigate the issue of improving solution quality by moving from a SAT to an ILP encoding in the context of a popular benchmark, the logistics planning domain (Veloso 1992). The scenario is the transportation of a set of packages that involves flights and truck drives between locations. There are various possible criteria to optimize in this domain: a) The sequential length, b) the parallel length, c) some function of the sequential and parallel lengths, and (d) ....

Veloso, M. 1992. Learning by analogical reasoning in general problem solving. Ph.D. Dissertation, CMU.

....weights and their scales play in the planning process, i.e. how they affect the planning time and the quality of the resulting plans. This is performed through an adequate number of experiments in an enhanced logistics type domain 1 . 5. 1 The logistics MO Domain The original logistics domain [29] consists of several locat ions in several cit ies. One location in each city is characterized as airport. Each city has one or more trucks, which can move between the locations of the city. There are also several planes, that can fly between the airports. Finally, there are some packages that ....

M. Veloso, Learning by analogical reasoning in general problem solving, Ph.D. Thesis, Department of Computer Science, Carnegie Mellon University, 1992.

....CRAV01 BUCA01 NTE01 resource allocation, process design (generation of contingent plans) elaboration of process instances at instantiation and during execution, monitoring [20] for (and anticipation of) deviations from plans, and re planning in the event of such exceptions. Learning techniques [21] could also be applied, allowing e.g. optimisation of processes over time. In this paper we have focussed on how the contingent planner, Cassandra, can help to automate the design of appropriate templates in a legacy system used to support the business processes at BT. ACKNOWLEDGEMENTS The first ....

Manuela Veloso. Learning by Analogical Reasoning in General problem solving. PhD thesis, Carnegie Mellon University, Pittsburgh, PA, 1992.

....found during past search. Case based reasoning is well adapted to problems where experience extensively represents domain knowledge and where only a weak domain theory is available [13] Control learning for planning has been investigated in machine learning as well as in case based reasoning [11, 16, 22]. Optimisation differs from planning in nature: First, planning focuses on how to achieve solutions, while in local search optimisation, where solutions are usually constructed easily, the task is to find good ones [10, 23] As consequence, memorising operator chainings, i.e. receipts for ....

.... matching complexity in case there is only one object type, as may happen in simple optimisation domains like single machine scheduling [10] Another common technique for the complexity reduction of source target matching, by reducing the length of the retained state description, is footprinting [22]. As in ALOIS a case focuses on only one single operator, footprinting only produces the operator s instanciated precondition. This is a somewhat restrictive description of the situation the operator was applied in. A case should ideally apprehend a description, as complete as possible, of the ....

Veloso M., Learning by analogical reasoning in general problem solving, Ph.D. thesis, Carnegie Mellon University, Pittsburgh, 1992

....handled efficiently around 10 6 states) in less than 10 seconds is a very encouraging result. Even if comparable with graphplan on the fixit problem, mbp seems in general to decrease performance in domains which can be solved by plans with parallel actions. We have performed a test on logistic [ 18 ] , a domain where most of the actions can be performed in parallel. logistic.a and logistic.b are two examples with different numbers of cities and packages (5 and 4, and 3 and 6, respectively) These examples are rather hard. ucpop was unable to find a solution even for problems with 3 cities and ....

Manuela Veloso. Learning by analogical reasoning in general problem solving. PhD thesis, 1992. CMU, CS Techn. Report CMU-CS-92-174.

....most of the work, at least in the domains we considered. An interesting research question is whether there is a more e#ective way to encode rules such as rule #3. Empirical Evaluation The testbed used in this paper is a series of problems from the logistics planning domain and the rocket domain (Veloso 1992; Blum and Furst 1995; Kautz and Selman 1996; Mcdermott 1998) and the tire world domain from the TLPlan distribution (Bacchus and Kabanza 1998) In addition, we created two new problem instances: logistics 1 and logistics 2, which can be solved with highly parallel plans (up to 20 actions in ....

....learning techniques for acquiring control knowledge automatically by training the planner on a sequence of smaller problems. Learning of control knowledge has been explored previously for other, more procedural, planners. See, for example, Etzioni (1993) Knoblock (1994) Minton (1988) and Veloso (1992). We are currently exploring forms of control rule learning in our declarative constraint based framework. Acknowledgements We thank Fahiem Bacchus, Carla Gomes, Dana Nau, and Dan Weld for many useful comments and suggestions. The second author received support from by an NSF Career grant and a ....

Veloso, M. (1992). Learning by analogical reasoning in general problem solving. Ph.D. Thesis, CMU, CS Techn. Report CMU-CS-92-174.

.... AI CBR viewpoint because, while it is similar to various AI CBR models (e.g. HYPO in comparing cases along factors [Ashley, 1990] PROTOS in comparing problems to prototypical or paradigmatic cases [Bareiss, 1989] Veloso s PRODIGY ANALOGY in comparing cases to select appropriate actions to apply [Veloso, 1992]) it introduces important components of CBR that have not yet been modeled such as: 1) symbolically comparing problems and paradigmatic cases to resolve conflicts among applicable general principles and (2) adequately accounting for a problem s specific contextual circumstances in deciding how ....

Veloso, M. M. (1992) Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie Mellon University. Technical Report No. CMU-CS-92-174.

.... all plans for getting an object between locations in the same city Q5: find all plans for getting an object between locations in different cities Q6: find all top level plans where a truck and an object are in the same location in the initial situation 10 This domain was based on that used in [15]. 12 For example, one query (Q2) was to find all plans using a particular airport. This query might be useful in avoiding plans that used an airport that is closed in the current situation. The sample transport logistics queries are considerably more structurally and computationally complex than ....

M.M. Veloso, Learning by Analogical Reasoning in General Problem Solving, Ph.D. disseration, CMU-CS-92-174, Dept. of Computer Science, Carnegie Mellon University, 1992.

....work has proposed an alternative derivational analogy in which the planner stores the choices made during the original planning episode and can more easily determine whether the same choices are applicable to the new situation. Derivational analogy has been used with both total order planners (Veloso 1992) and partial order planners (Ihrig Kambhampati 1994) Hierarchical Matching Analogy works well if the goal states and initial states are relatively small (even though the plans themselves may be quite complex) However, if most problems have several goals the planner must simultaneously ....

Veloso, M. 1992. Learning by Analogical Reasoning in General Problem Solving. Ph.D. Dissertation, Carnegie Mellon.

....we describe the conditions under which it is caused not to occur whenever PickUp(B) occurs, for some block B. These auxiliary inaction atoms are used to stipulate that a pickup action occurs if and only if a putat action does. 7. 2 LOGISTICS PLANNING PROBLEMS The logistics domain is due to Veloso [1992]. Kautz and Selman studied three large logistics planning problems. Due to space constraints, we do not describe them, nor do we display an example input file. The logistics domain is more complex than the blocks world domain. It includes several kinds of actions that can occur concurrently. Our ....

Manuela Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, CMU, 1992. CS Technical Report CMU-CS-92-174.

....this domain knowledge by cases (Kolodner, 1993) They are appropriate for domains where a strong theory does not exist but past experience is accessible. A plan is a specific sequence of steps (or actions) with the aim of a goal achievement. Case Based Planning (CBP) systems (Hammond, 1989; Veloso, 1992) reuse past sequences of actions from past plans to construct new ones. There are some case based models for the creative process. Linda Wills and Janet Kolodner (Wills Kolodner, 1994) consider three steps in creative design: enumeration of several alternative solutions; re description and ....

Veloso, M., (1992). Learning by Analogical Reasoning in General Problem Solving, PhD D thesis, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA.

....(style style stir fry) Figure 2.4: CHEF s recipe for broccoli with tofu. Two recent planners have added case based reasoning to nonlinear planners and taken the reasoning process by which the plan was generated (i.e. the final successful search path through the plan space) as their cases [Veloso, 1992, Kambhampati and Hendler, 1989] When solving a new problem, they adapt the method used to solve a previous problem, rather than the previous problem s solution. This approach, known as derivational analogy, is similar to the way in which a student will solve a calculus problem, or construct an ....

Veloso, Manuela M. 1992. Learning by analogical reasoning in general problem solving. Technical Report CMU-CS-92-174, School of Computer Science, Carnegie Mellon University. (PhD Thesis).

....sensitivity of asp to different time windows and perturbations, and end with a summary of the main results and topics for future work. Preview of Results In our experiments we focused on the domains used by Kautz and Selman (1996) the rocket domain (Blum Furst 1995) the logistics domain (Veloso 1992), and the blocks world domain. Blum s and Furst s graphplan outperforms prodigy (Carbonell et al. 1992) and ucpop (Penberthy Weld 1992) on the rocket domains, while satplan outperforms graphplan in all domains by at least an order of magnitude. Table 1 compares the performance of the new ....

Veloso, M. 1992. Learning by Analogical Reasoning in General Problem Solving. Ph.D. Dissertation, Computer Science Department, CMU. Tech. Report CMU-CS-92174.

....and without the new splitting ability. As handling large domains is one of the important motivations for using HTN planners, we performed the tests on a set of problems from the UM Translog domain (Andrews et al. 1995) The UM Translog domain was inspired by the CMU logisticstransportation domain (Veloso 1992), but since UM Translog deals with various types of packages, vehicles, and procedures for handling the packages, it is an order of magnitude larger the CMU logistics. The UM Translog domain is specified in HTN using 41 operators and 45 decomposition methods. Typical problems in that domain deal ....

Veloso, M. 1992. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, School of Computer Science, Carnegie Mellon University. Technical report CMU-CS-92-174.

.... direct use of specially formulated adaptation knowledge to determine simple surface changes, structural transformations, and complex interactions [6] our integration of adaptation knowledge into retrieval is influenced by techniques used to introduce other forms of domain knowledge into retrieval [7,8]. Furthermore, AGR works without incurring the full cost of adaptation during retrieval. Indeed, AGR can be more efficient than standard methods in CBR. AGR is implemented in Dj Vu, a case based reasoning system for software design (see section 2 and 3) In section 4, we show how ....

....because the cost of retrieval is directly proportional to the number of cases in the case base; as a case base expands overall problem solving performance may actually degrade. One solution to this problem is to limit retrieval time; the best case found within a given time limit is selected [8]. This solution invariably results in the retrieval of a sub optimal case and, of course, such sub optimal cases may be difficult or impossible to adapt. Adaptation guided retrieval is less prone to the swamping problem, because the cost of retrieval does not depend on the size of the case base as ....

Veloso, M. (1992) Learning by Analogical Reasoning in General Problem Solving. Ph.D. Thesis (CMU-CS-92-174). Carnegie Mellon University, Pittsburgh, USA.

.... do we formulate the learning task in the context of multi agent interactions where procedural and control knowledge must be learned Concept learning has been the focus of most machine learning research (e.g. 10] Learning of control knowledge has been explored using case based reasoning (e.g. [8, 20, 11]) and reinforcement type learning techniques (e.g. 14, 12] This research has been conducted almost exclusively in a single agent setting. We want to explore strategies for multiple agent learning of control knowledge during agent interactions. Within each formulation of the learning task ....

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, School of Computer Science, Carnegie Mellon University, 1992. 14

....exponent may exist for finding exact or approximate solutions. Research in machine learning have long studied the problem of automatically creating e#cient planners by learning domain specific rules or cases to control a general search engine (Minton, 1988; Carbonell, Knoblock, Minton, 1990; Veloso, 1992; Etzioni, 1993; Bhatnagar Mostow, 1994; Kambhampati, Katukam, Qu, 1996; Borrajo Veloso, 1997; Aler, Borrajo, Isasi, 1998; Leckie Zukerman, 1998; etc. However, the successful practical application of machine learning techniques has been limited by at least two factors: First, ....

.... Pazzani, 1991) Recently Kambhampati (1999) has shown how explanation based learning (EBL) techniques can be applied to Graphplan, but does not attempt to learn high level, declarative rules. We will illustrate the details of the system using examples from a logistics planning domain (Veloso, 1992) which has appeared as a benchmark in most recent work in planning. In brief, the task in this domain is to move a set of packages from various initial locations to various goal locations. Packages can be moved between locations within a city by truck. Airplanes can transport packages between ....

Veloso, M. M. (1992). Learning by analogical reasoning in general problem solving . Doctoral dissertation, Department of Computer Science, Carnegie Mellon University, Pittsburgh.

....by many real world tasks, the efficiency benefit is brought into question. Specifically, a problem known as the utility problem means that efficiency will eventually degrade as the case base becomes saturated . The utility problem has been well documented by the machine learning community [1,2,3,4]. Evidently, large amounts of training data will affect the performance of the training process. However, in eager learning systems, the competence of a system can deteriorate with large amounts of training data due to over fitting to the training data for instance. One of the advantages of the ....

Veloso, M., Learning by Analogical Reasoning in General Problem Solving. Ph.D Thesis (CMU-CS-92-174). Carnegie Mellon University, Pittsburgh, USA, (1992).

.... useful and reliable results, they need to have reasonably sized case knowledge bases (CKB s) For some domains this is not much of a problem since the representations are either fairly simple (Golding Rosenbloom 1991; Lehnert 1987b) or can be easily automatically generated (Lehnert 1987a; Veloso 1992). However, when the domain representation for a case is complex, building a case knowledge base with large numbers of cases typically becomes an expensive proposition. Converting case information into a symbolic representation will frequently make use of the time and experiences of a subject ....

Veloso, M. M. 1992. Learning by Analogical Reasoning in General Problem Solving. Ph.D. Dissertation, Carnegie Mellon University, Pittsburgh, PA.

....been done in the past. In order for CBR systems to generate useful and reliable results, they need to have reasonably sized case knowledge bases (CKB s) For some domains this is not a problem since the representations are either fairly simple [7, 13] or can be easily automatically generated [12, 16]. However, when the domain representation for a case is complex, it is typically the situation that building a CKB with large numbers of cases becomes an expensive proposition. Converting information into a symbolic case representation will frequently make use of the time and experiences of a ....

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie Mellon University, Pittsburgh, PA, August 1992. 7

....implements the algorithm for learning relational strategies (without support predicate and internal state) as in Corollary 5.6. This algorithm is applied to small planning domains that have been studied before, including a four operator version of the blocks world, and the logistics domain (Veloso, 1992). The experiments demonstrate that our results are indeed applicable, that rule based strategies are useful for such domains, and that the algorithm is 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 1000 2000 3000 4000 5000 6000 Fraction of problems solved Cumulative run length 7 blocks 8 blocks 10 blocks ....

Veloso, M. (1992). Learning by Analogical Reasoning in General Problem Solving. Ph.D.

....[24, 27, 14, 47, 2] Typically these approaches are based around a search engine, and encode control knowledge in some way so as to direct the search mechanism. A variety of methods to extracting control knowledge including Explanation Based Learning [24, 27] static analysis [14] and analogy [45, 46] have been used in this approach, and indeed show some success in several planning domains. Another response to the difficulty of the problem is to abandon search altogether and construct a special algorithm for a planning domain. This algorithm can be thought of as a mapping from any situation ....

....the domains studied here such strategies can be found. We have experimented with two domains: The blocks world domain has been widely studied before, and due to recent studies [18, 8, 37, 38] its structure is well understood so as to enable thorough analysis. The Logistics transportation domain [45] is more complex and has been recently studied from several perspectives [45, 47, 13, 20] For each domain, random problems are drawn and presented with their solutions to the learning algorithm. The learning system uses a variant of Rivest s algorithm [33] to produce a PRS strategy represented as ....

[Article contains additional citation context not shown here]

M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, School of Computer Science, Carnegie Mellon University, 1992. Also appeared as technical report CMU-CS-92-174.

....in section 3. Since we do not make any fundamental changes to the planner, we are able to take advantage of the body of work that has been done with classical planners, such as the use of abstraction [18] machine learning to improve planning performance [21, 11, 15] and derivational analogy [27, 16]. To give a feel for the type of behavior we have been able to get from our architecture, in section 4 we provide two traces of the system controlling a simulated household robot built in the Oz system [1] In section 5 we present the results of some experiments we ran in the household robot ....

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie Mellon University, august 1992.

....in search) When planning and learning are interleaved and the planner stores knowledge that will be useful later, a long term use of space has to be considered. The stored knowledge can take the form of search control rules extracted from problem solving traces, or of cases in a case library [Veloso, 1992, Kambhampati, 1990, Hammond, 1986] Recycling past successful experience reduces the search effort when solving new similar problems. Note that there is usually a trade off among the amount of knowledge stored, the cost of accessing and reusing it, and the savings on search gained from it ....

....required to generate plans. In many domains finding a plan at all requires a considerable amount of search and there has been work on improving the efficiency of a problem solver with machine learning techniques [Mitchell et al. 1983, Laird et al. 1986, Gratch and DeJong, 1992, Korf, 1985, Veloso, 1992, Minton, 1988, Knoblock, 1991b, Etzioni, 1990] We call this speed up learning. However these mechanisms have paid none or little attention to the quality of the solutions obtained. Here we briefly present some examples of speed up learning systems in the context of PRODIGY. PRODIGY s ....

[Article contains additional citation context not shown here]

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie Mellon University, School of Computer Science, August 1992. Available as technical report CMU-CS-92-174.

....of rules grows and covers increasing amounts of the input space. This will, since the sequential evaluation of a network is generally slow, speed up the classification of unknown instances an effect that has been studied extensively in the context of symbolic problem solvers [Minton, 1988] V eloso, 1992] However, there is a trade off, since as the corpus of rules becomes too large, searching through the rules may exceed the time required for network evaluation. In a series of experiments we collected rules generated from random queries. The goal was Extracting Symbolic Knowledge from ....

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie Mellon University, School of Computer Science, Pittsburgh, PA, August 1992.

....available cases [1] Such general knowledge may state dependencies between certain case features and can be used to infer additional, previously unknown features from the known ones. Furthermore, some applications require an adaptation of a retrieved case according to the actual problem at hand [13, 6, 4]. Therefore, general knowledge is required to specify such an adaptation. This paper addresses the representation and the processing of background knowledge required for case based reasoning applications in the field of classification, diagnosis, and decision support with the INRECA system. This ....

....it is known in advance that they can be adapted by the available adaptation rules. Moreover, our approach is limited to transformational adaptation, which is sufficient for most classification and diagnosis applications. Currently, neither knowledge required to perform derivational adaptation [5, 13] (as required for planning tasks) nor knowledge which is naturally expressed as a set of constraints (as required for design task) can be represented and processed within our approach. Acknowledgements The authors want to thank Ivo Vollrath for the realization of this work and Harald Holz and Ivo ....

M. M. Veloso, Learning by analogical reasoning in general problem solving, Ph.D. dissertation, Carnegie Mellon University, Pittsburgh, PA, 1992.

....together with a related solution. As is the case in Prodigy (Minton et al. 1989) we only consider sequential plans, i.e. plans with totally ordered operators. The planning cases we assume do not include a problem solving trace as for example the problem solving cases in Prodigy Analogy (Veloso, 1992; Veloso Carbonell, 1993; Veloso, 1994) In real world applications, a domain expert s solutions to previous problems are usually recorded in a company s filing cabinet or database. These cases can be seen as a collection of the company s experience, from which we want to draw power. During a ....

.... of a plan, the plans robustness, or certain user preferences (Perez Carbonell, 1993) Because such quality measures are very difficult to assess, in particular in our manufacturing domain, we rely on this simple criterion also used for evaluating the quality of solutions in Prodigy Analogy (Veloso, 1992). 9.5.1 Experimental Setting We have analyzed the solutions computed in the previous set of experiments to assess the quality of the solutions produced by Paris. Therefore, the length of solutions derived during problem solving, after learning from each of the 20 training sets, are compared to ....

Veloso, M. M. (1992). Learning by analogical reasoning in general problem solving. Ph.D.

....for finding near optimal solutions to problems for which feasible solutions can be constructed easily. In both fields, there is increasing interest in finding near optimal solutions to problems with a difficult feasibility aspect. Here, we consider a planning benchmark from the logistics domain (Veloso 1992) that also exhibits this characteristic. The scenario is the transportation of a set of packages that involves flights and truck drives between locations. To model the problem, we use a variant of the state based encodings presented in (Kautz Selman 1996) that is extended to encode a notion of ....

Veloso, M. 1992. Learning by analogical reasoning in general problem solving. Ph.D. Dissertation, CMU.

No context found.

Veloso, M. M. 1992. Learning by Analogical Reasoning in General Problem Solving. Ph.D. Dissertation, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA. Available as technical report CMU-CS-92-174. A revised version of this manuscript will be published by Springer Verlag, 1994.

....efficiency of the planner. For example, Prodigy is able to learn control rules [Minton, 1988] conduct experiments to acquire new knowledge [Gil, 1992] generate abstraction hierarchies [Knoblock, 1993] and use andlogical reasoning to recognize and exploit similarities between planning problems [Veloso, 1992]. Prodigy s core, the planning algorithm itself, has been improved over the years. The old algorithm, Prodigy2.0 [Minton et al. 1989] was succeeded by NoLimit [Veloso, 1989] and then by Prodigy4.0 [Carbonell et al. 1992] All versions of Prodigy were developed by members of the Prodigy ....

....the efficiency of the Prodigy system. A formal analysis of advantages and drawbacks of Prodigy as compared to other planners is still an open research problem. However, multiple experiments have demonstrated Prodigy s ability to efficiently solve a wide range of complex problems (see, for example, [Veloso, 1992] and [Gil, 1992] Below we list some advantages of Prodigy s planning algorithm. Rich domain language. Prodigy s language for operator representation includes dis 9 junctive preconditions, universal and existential quantification of variables, and conditional and functional effects. The ....

Manuela Veloso. Learning by analogical reasoning in general problem solving. PhD thesis, Carnegie Mellon, Computer Sci., 1992. Tech. Rep. CMU-CS-92-174.

....the library have demonstrated the scaling properties of the memory organization, of the match retrieval process, and of the reconstruction mechanism replaying multiple cases. Details on the multiple case replay mechanism and results of these scal ing up tests are being compiled and forthcoming in [Veloso, 1992]. We now show two examples from the logistics domain that illustrate the replay mechanism when one or more cases are used for guidance. 6.1 Following one case Subgoaling structure and failures In Figure 14 we illustrate how the subgoaling structure and the failure records at a stored case can ....

....two situations becomes clearer along the reconstruction. In the figure we show the case further instantiated with the substitutions tr35 tr0 and p13 p10 that occur dynamically at transfer time as we now also explain. 5. The complete storage, matching, and retrieval procedures are described in [Veloso, 1992]. Figure 14. Following one case Subgoaling structure and failures. The same goal is chosen at the node nl as it was at node cnl. At node cn2 the past case records that the operator (unload airplane oh0 pl3 a2) was successfully chosen and that the operator (unload truck oh0 tr77 a2) failed. ....

[Article contains additional citation context not shown here]

Veloso, M. M. (1992). Learning by Analogical Reasoning in General Problem Solving. PhD thesis, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA. Available as technical report CMU-CS-92-174.

....chain responsible for a success or failure and compile search control rules therefrom. Similar efforts within the linear planner of prodigy were done to learn control rules from partially evaluating the domain theory [ Etzioni, 1990, P erez and Etzioni, 1992 ] In the nonlinear planner, Veloso, 1992 ] develops a case based learning method that consists of storing individual problems solved to guide the planner when solving similar new problems. The guiding similar plans provide global control knowledge in the sense that they consists of a chain of decisions. This control guidance contrasts ....

....as well as to other related problems. The rule allows the nonlinear problem solver to use the correct goal interleaving leading to the optimal solution. 4 Empirical Results We have been performing extended empirical experiments in the blocksworld and in the logistics transportation domain [ Veloso, 1992 ] We report here the results from initial experiments in the blocksworld that we fully analyzed and interpreted. The results illustrate our main claims about the effectiveness of the combined bounded explanation and inductive methods in speedup learning. First, we used a set of 50 problems ....

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA, August 1992. Available as technical report CMU-CS-92-174.

....logistics domain where we illustrate the phases of the generation of the control rules and their inductive refinement. In this domain, packages are to be moved among different cities. Packages are carried within the same city in trucks and across cities in This domain was first introduced in [23]. airplanes. At each city, there are several locations, e.g. post offices and airports. This transportation domain represents a considerable scale up in length of the solution, size of the search space, and other difficult learning issues, such as nonlinearity, un optimality of solutions, and a ....

....that is needed in order to deal with new problems effectively and with quality. This learning possibility is not encountered by a linear problem solver who handles multiple goals independently. It would also not be compiled as a local choice by other learning methods applied to nonlinear planning [12, 23]. The preconditions are the initial bounded explanation of the problem solving decision. This rule is over specific, since, among other things, it records the positions of other trucks available and another object present at the post office, which turn out to be irrelevant features for this ....

[Article contains additional citation context not shown here]

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA, August 1992. Available as technical report CMU-CS-92-174.

No context found.

M.Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie Mellon University. 1992.

No context found.

M. Veloso. Learning by analogical reasoning in general problem solving. PhD thesis, Carnegie-Mellon University, 1992.

No context found.

Manuela M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie-Mellon University, 1992. (CMU-CS-92).

No context found.

M. Veloso. Learning by analogical reasoning in general problem solving. PhD thesis, Carnegie-Mellon University, 1992.

No context found.

M. Veloso. Learning by analogical reasoning in general problem solving. PhD thesis, Carnegie-Mellon University, 1992.

No context found.

Veloso, M., 1992, Learning by Analogical Reasoning in General Problem Solving, Ph.D thesis, CMU. -20-

No context found.

Veloso, M. (1992) Learning by Analogical Reasoning in General Problem Solving. Ph.D. Thesis, (CMU-CS-92-174), School of Computer Science, Carnegie Mellon University, Pittsburgh, USA.

No context found.

Veloso M. (1992). Learning by Analogical Reasoning in General Problem Solving, Ph.D. Thesis, CMU-CS-92-174, School of Computer Science, Carnegie Mellon University, Pittsburgh, USA.

No context found.

M.M. Veloso. Learning by Analogical Reasoning in General Problem Solving. PhD thesis, Carnegie Mellon University, CMU, Pittsburgh, USA, 1992. CMU-CS-92-174.

First 50 documents  Next 50

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