基于神经网络非线性系统辨识和控制的研究

本文提出了由静态的前馈网络和稳定的滤波器构成的非线性系统的辨识模型，在神经网络固有的逼近误差存在的情况下，从理论上讨论了神经网络应用于辨识控制过程中系统的稳定性问题，最后研究了在非线性系统的轨迹跟踪过程中增加滑动控制来偿神经网络的逼近误差，从而提高系统跟踪性能。     

Technical Note: Selecting a Classification Method by Cross-Validation

If we lack relevant problem-specific knowledge, cross-validation methods may be used to select a classification method empirically. We examine this idea here to show in what senses cross-validation does and does not solve the selection problem. As illustrated empirically, cross-validation may lead to higher average performance than application of any single classification strategy, and it also cuts the risk of poor performance. On the other hand, cross-validation is no more or less a form of bias than simpler strategies, and applying it appropriately ultimately depends in the same way on prior knowledge. In fact, cross-validation may be seen as a way of applying partial information about the applicability of alternative classification strategies.     

Perceptually Grounded Language Learning: Part 2—DETE: A Neural/Procedural Model

In Part 1 of this two-part series, we introduced Katamic memory—a neural network architecture capable of robust sequence learning and recognition. In Part 2, we introduce the Blobs World taskjdomain for language learning and describe the DETE language learning system, which is composed of over 50 Katamic memory modules. DETE currently learns small subsets of English and Spanish via association with perceptual! motor inputs. In addition to Kaiamic memory, DETE employs several other novel features: (1) use of feature planes, to encode visual shapes, spatial relationships and the motions of objects, (2) phase-locking of neural firing, in order to represent focus of atention and to bind objects across multiple feature planes, and (3) a method for encoding temporal relationships, so that DETE can learn utterances involving the immediate past and future. We compare DETE to related models and discuss the implications of this approach for language-learning research.     

Artificial intelligence and symbols

The introduction of massive parallelism and the renewed interest in neural networks gives a new need to evaluate the relationship of symbolic processing and artificial intelligence. The physical symbol hypothesis has encountered many difficulties coping with human concepts and common sense. Expert systems are showing more promise for the early stages of learning than for real expertise. There is a need to evaluate more fully the inherent limitations of symbol systems and the potential for programming compared with training. This can give more realistic goals for symbolic systems, particularly those based on logical foundations.     

Environmentally friendly, heterogeneous acid and base catalysis for the methylation of catechol: Chances for the control of chemo-selectivity

The properties of catalysts with (i) Brønsted-type acidity (H-mordenite and Al/P mixed oxide), (ii) Lewis-type acidity (Al trifluoride) or (iii) basic characteristics (Mg/Fe mixed oxide) were investigated in the gas-phase methylation of catechol. When methanol was used as the methylating agent, H-mordenite and AlF₃ gave high selectivities to guaiacol (the product of O-methylation) under mild reaction conditions, that is at very low catechol conversions. An increase in temperature led to the transformation of guaiacol to phenol and cresols, and to considerable catalyst deactivation. The basic catalyst Mg/Fe/O also favored an extensive degradation of guaiacol to phenol. On the mildly acidic catalyst Al/P mixed oxide a stable catalytic performance and a high selectivity to guaiacol at 40% catechol conversion were obtained. When methylformate, a more reactive methylating agent, was used with AlF₃ and Mg/Fe mixed oxide as catalysts, higher catechol conversions and slower deactivation rates could be achieved under mild reaction conditions, with a low extent of guaiacol degradation. However, methylformate rapidly decomposed when temperatures above 350 °C were used. Finally, tests were made by reacting catechol and diethoxymethane with acid catalysts, with the aim of synthesizing methylenedioxybenzene. The latter product was obtained with high selectivity, but with very low yield, due to both catalyst deactivation and decomposition of diethoxymethane.     

An expert decision support system for network routing: a case study

The paper presents a case study of the development of an expert decision support system which uses simple heuristic methods for fast determination of routes for simultaneous signals in a transmission network of limited capacity. It illustrates how heuristic solutions can be embodied in a model-based DSS and how the standard decision support literature, although intuitively appealing, provides little practical assistance in system construction or classification     

Evaluation of rainfall networks using entropy: II. Application

This paper, the second in the series, applies the entropy approach to space and time evaluation of rainfall networks in Louisiana. The evaluation is made for five sampling intervals: daily, 2-day, weekly, monthly, and yearly, and for two separate seasons. In each case, the best combination of raingages is suggested and the lines of equal information (isoinformation contours) are constructed showing the areas of greater or less information transfer. The isoinformation lines are especially suitable in decisions concerning the expansion of the existing network or deleting the unnecessary raingages.     

Optimized Schwarz methods without overlap for highly heterogeneous media

In this paper an original variant of the Schwarz domain decomposition method is introduced for heterogeneous media. This method uses new optimized interface conditions specially designed to take into account the heterogeneity between the sub-domains on each sides of the interfaces. Numerical experiments illustrate the dependency of the proposed method with respect to several parameters, and confirm the robustness and efficiency of this method based on such optimized interface conditions. Several mesh partitions taking into account multiple cross points are considered in these experiments.     

Heterogeneous object modeling: A review

A review on the recent development in heterogeneous object modeling is provided in this paper. General problems and prevalent solutions to the modeling of heterogeneous objects are investigated. Vigorous heterogeneous object representations are roughly classified into three major categories: evaluated models, unevaluated models and composite models. We reveal their similarities and study their strengths and weaknesses in terms of the representational capacities, intuitiveness, exactness, compactness, efficiency and other criteria. Different design paradigms are briefly reviewed to provide an overview of the state-of-the-art in heterogeneous object design. Finally, open problems and possible future directions are discussed.     

Reasoning about non-immediate triggers in biological networks

Modeling molecular interactions in signalling networks is important from various perspectives such as predicting side effects of drugs, explaining unusual cellular behavior and drug and therapy design. Various formal languages have been proposed for representing and reasoning about molecular interactions. The interactions are modeled as triggered events in most of the approaches. The triggering of events is assumed to be immediate: once an interaction is triggered, it should occur immediately. Although working well for engineering systems, this assumption poses a serious problem in modeling biological systems. Our knowledge about biological systems is inherently incomplete, thus molecular interactions are constantly elaborated and refined at different granularity of abstraction. The model of immediate triggers can not consistently deal with this refinement. In this paper we propose an action language to address this problem. We show that the language allows for refinements of biological knowledge, although at a higher cost in terms of complexity.