![]() |
|||
![]()
|
![]() |
![]() |
![]() |
4.4. DECISION TREES AND CASE-BASED REASONINGBoth machine learning techniques take a set of training examples/cases as input. Each training example/case consists of a set of decision choices and a corresponding class. The decision tree approach constructs a decision tree that has internal nodes labeled as decisions and the leaves labeled as classes. At run time, a classification is done by following a path from the root node to a leaf, given only a set of decision choices. In contrast, the case-based reasoning approach just stores training cases in a memory. It classifies a new case by comparing it with stored cases at run time using a variety of indexing and matching strategies. The class of the most similar case is determined as the classification. These two machine learning techniques can be used to acquire domain knowledge through collecting training examples, and complement the direct approach of rule acquisition. Unlike neural networks, their reasoning processes can be analyzed symbolically. Note that each technique has its own specific classification bias (e.g., the decision tree approach is sensitive to the order of decision choices). Decision tree and case-based reasoning have been utilized to analyze data and construct domain expertise incrementally in telecommunications domains. Their tasks include alarm diagnosis, traffic control, workflow management, customer helpdesk, churning management, and data mining. 4.5. MODEL-BASED REASONINGModel-based reasoning is often used in the telecommunications domains where the dynamics or behavior can be modeled. Inference based on a correct model can be deep, direct, and efficient. The representation of such a model can be tables, finite machines, semantic networks, logic formulas, or Hidden Markov models. Typically, given the observed input, model-based systems use a model to predict the expected behavior, compare it with actual behavior, and then proceed to the next step in the model or stop to draw some conclusion. If a model is available, the correctness and completeness of the model determines the quality of model-based reasoning. Model-based reasoning has been traditionally used for well-defined physical systems, yet it has also found its place in alarm correlation, network monitoring, configuration management, language translation, and speech recognition. 4.6. DISTRIBUTED ARTIFICIAL INTELLIGENCEDistributed Artificial Intelligence (DAI) addresses cooperative and distributed problem-solving. Its paradigms include contracting, blackboard systems, distributed search, speech-act communication, and agent-based belief systems. DAI's goal is to find a globally acceptable solution from distributed and often limited local systems through their communication and coordination. Research in DAI involves studying the emerging behavior and convergence property of distributed agents. It also investigates the trade-off between problem solving and communication. DAI adds distribution to the complexity of AI technology. Telecommunications domains are often distributed along spatial, functional, and organizational dimensions. In the past, special-purpose expert systems were developed independently. Recently, there is a trend to push toward the convergence and integration of these systems. For example, cooperation of management tasks is needed not only within the same management layer but also across all four layers. Public and private networks should be managed from both a logical network perspective and a physical network perspective. DAI techniques can be used to glue different expert systems (Velthuijsen, 1996). They have been tried in fault isolation, network design and management, dynamic channel allocation, resource scheduling, traffic control and routing, service order, and workflow management. 4.7. APPROXIMATE REASONINGApproximate reasoning deals with reasoning under uncertainty. This class of techniques is based on statistics, probability theory, fuzzy logic, or decision theory. They conduct reasoning processes using numbers rather than symbols, and their reasoning relies on a large amount of data, known prior beliefs, or some likelihood distribution. Their approaches range from combining rules using likelihoods or fuzzy logic to using utility-based decisions or Bayesian networks. They are often used to enhance the robustness of logic-based techniques and their conclusions, by nature, include uncertainty. Due to the imprecise nature of many telecommunications domains and poor quality of information in databases, approximate reasoning has been found useful for fault isolation, trending analysis, data mining, billing, and growth planning. 4.8. HYBRID SYSTEMSEach of above AI techniques has certain strengths and weaknesses. In practice, two or more of them are often used together to complement each other. Examples include combining rule-based systems and neural networks (Tan, 1996), rule-based and model-based systems (Worrest, 1996), search and DAI (Low, 1995), model-based systems and approximate reasoning (Chen, 1996), and case-based reasoning and decision trees (Masand, 1996). They can be loosely coupled (e.g., one's output is other's input), tightly coupled (e.g., blackboard paradigm allows different AI techniques to communicate with each other during problem-solving), or fully integrated (e.g., fuzzy rule-based systems). Keep in mind that no one AI technique is effective in all domains. Every AI technique has its respective domains where it is effective. Again, the key is to find a good match between AI techniques and a specific domain. 5. APPLICATIONSToday, telecommunications networks are highly advanced, rapidly evolving systems of complex, interdependent technologies. As telecommunications networks fuse with the Internet, and as the underlying technologies continue their rapid evolution, these networks will become increasingly difficult to manage. AI is playing a large and growing role in various telecommunications management tasks. This section describes some of fielded AI applications in wireline and wireless communications.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |