004 Informatik
Refine
Year of publication
Document Type
- Report (22)
- Working Paper (11)
- Article (7)
- Preprint (5)
- Conference Proceeding (2)
Has Fulltext
- yes (47)
Keywords
- Optimierung (17)
- Optimization (12)
- Modellierung (9)
- Soft Computing (8)
- Globale Optimierung (5)
- Simulation (5)
- Evolutionärer Algorithmus (4)
- Metaheuristik (4)
- Modeling (4)
- Benchmarking (3)
Learning board games by self-play has a long tradition in computational intelligence for games. Based on Tesauro’s seminal success with TD-Gammon in 1994, many successful agents use temporal difference learning today. But in order to be successful with temporal difference learning on game tasks, often a careful selection of features and a large number of training games is necessary. Even for board games of moderate complexity like Connect-4, we found in previous work that a very rich initial feature set and several millions of game plays are required. In this work we investigate different approaches of online-adaptable learning rates like Incremental Delta Bar Delta (IDBD) or Temporal Coherence Learning (TCL) whether they have the potential to speed up learning for such a complex task. We propose a new variant of TCL with geometric step size changes. We compare those algorithms with several other state-of-the-art learning rate adaptation algorithms and perform a case study on the sensitivity with respect to their meta parameters. We show that in this set of learning algorithms those with geometric step size changes outperform those other algorithms with constant step size changes. Algorithms with nonlinear output functions are slightly better than linear ones. Algorithms with geometric step size changes learn faster by a factor of 4 as compared to previously published results on the task Connect-4.
Computational intelligence methods have gained importance in several real-world domains such as process optimization, system identification, data mining, or statistical quality control. Tools are missing, which determine the applicability of computational intelligence methods in these application domains in an objective manner. Statistics provide methods for comparing algorithms on certain data sets. In the past, several test suites were presented and considered as state of the art. However, there are several drawbacks of these test suites, namely: (i) problem instances are somehow artificial and have no direct link to real-world settings; (ii) since there is a fixed number of test instances, algorithms can be fitted or tuned to this specific and very limited set of test functions; (iii) statistical tools for comparisons of several algorithms on several test problem instances are relatively complex and not easily to analyze. We propose amethodology to overcome these dificulties. It is based on standard ideas from statistics: analysis of variance and its extension to mixed models. This work combines essential ideas from two approaches: problem generation and statistical analysis of computer experiments.
Computational intelligence methods have gained importance in several real-world domains such as process optimization, system identification, data mining, or statistical quality control. Tools are missing, which determine the applicability of computational intelligence methods in these application domains in an objective manner. Statistics provide methods for comparing algorithms on certain data sets. In the past, several test suites were presented and considered as state of the art. However, there are several drawbacks of these test suites, namely: (i) problem instances are somehow artificial and have no direct link to real-world settings; (ii) since there is a fixed number of test instances, algorithms can be fitted or tuned to this specific and very limited set of test functions; (iii) statistical tools for comparisons of several algorithms on several test problem instances are relatively complex and not easily to analyze. We propose a methodology to overcome these difficulties. It is based on standard ideas from statistics: analysis of variance and its extension to mixed models. This paper combines essential ideas from two approaches: problem generation and statistical analysis of computer experiments.
This paper proposes an experimental methodology for on-line machine learning algorithms, i.e., for algorithms that work on data that are available in a sequential order.
It is demonstrated how established tools from experimental algorithmics (EA) can be applied in the on-line or streaming data setting.
The massive on-line analysis (MOA) framework is used to perform the experiments.
Benefits of a well-defined report structure are discussed.
The application of methods from the EA community to on-line or streaming data is referred to as experimental algorithmics for streaming data (EADS).
This report presents a practical approach to stacked generalization in surrogate model based optimization. It exemplifies the integration of stacking methods into the surrogate model building process. First, a brief overview of the current state in surrogate model based opti- mization is presented. Stacked generalization is introduced as a promising ensemble surrogate modeling approach. Then two examples (the first is based on a real world application and the second on a set of artificial test functions) are presented. These examples clearly illustrate two properties of stacked generalization: (i) combining information from two poor performing models can result in a good performing model and (ii) even if the ensemble contains a good performing model, combining its information with information from poor performing models results in a relatively small performance decrease only.
Faserverbundwerkstoffe (FVW) und Composites haben in der Luft- und Raumfahrtindustrie, im Automobilbau, beim Bau von Windenergieanlagen und in vielen weiteren zukunftsträchtigen Branchen eine große Bedeutung. Maßnahmen, die ein Erkennen von Schädigungen simultan zur Entstehung ermöglichen und Restbetriebszeiten prognostizieren können, sind geeignet, die Lebensdauer von FVW-Konstruktionen zu erhöhen. Darüber hinaus ist eine zustandsorientierte und somit kosteneffektive Wartung dieser Bauteile möglich.
Sowohl die Prognose, als auch die Detektion von Schäden würde den ressourcenschonenden Einsatz dieser Werkstoff-gruppe ermöglichen. Das sogenannte Structural Health Monitoring (SHM) bezeichnet in diesem Zusammenhang eine Methode, die es ermöglicht, kontinuierlich Anhalts-punkte über die Funktionsfähigkeit von Bauteilen und Konstruktionen zu erhalten.
Dieser Artikel beschreibt die Planung, Durchführung und Analyse von SHM-Experimenten. Das Hauptziel bestand in der Planung von Experimenten zur Gewinnung von Messdaten mittels piezoelektrischen Elementen auf Versuchstafeln, bei denen bewusst trukturbeschädigungen eingebracht wurden. Statistische Auswertungsmethoden sollen auf ihre Eignung getestet werden, Rückschlüsse aus den experimentell gewonnenen Daten auf die Art der Strukturbeschädigungen zu ziehen.
Evolutionary algorithm (EA) is an umbrella term used to describe population-based stochastic direct search algorithms that in some sense mimic natural evolution. Prominent representatives of such algorithms are genetic algorithms, evolution strategies, evolutionary programming, and genetic programming. On the basis of the evolutionary cycle, similarities and differences between these algorithms are described. We briefly discuss how EAs can be adapted to work well in case of multiple objectives, and dynamic or noisy optimization problems. We look at the tuning of algorithms and present some recent developments coming from theory. Finally, typical applications of EAs to real-world problems are shown, with special emphasis on data-mining applications
Dieser Schlussbericht beschreibt die im Projekt „Methoden der Computational Intelligence für Vorhersagemodelle in der Finanzund Wasserwirtschaft“ (FIWA) im Zeitraum von Juni 2009 bis einschließlich November 2012 erzielten Ergebnisse. In der Praxis werden für diese Vorhersagemodelle Verfahren der linearen und nichtlinearen Regression, NN, Support Vector Machines (SVM) und viele weitere Verfahren eingesetzt. Das Projekt FIWA befasste sich mit der Entwicklung modularer Systeme zur Analyse und Prognose von Daten aus der Finanz- und Wasserwirtschaft mittels Verfahren der Computational Intelligence (CI) mit methodischem Fokus auf dem CI-Unterbereich Genetic Programming (GP). Ein zentrales Ergebnis der wissenschaftlichtechnischen Arbeit im Projekt FIWA ist die Entwicklung der Open-Source Software RGP. Dabei handelt es sich um ein Software- Framework für GP, welches auf die automatische Erstellung von Vorhersagemodellen spezialisiert ist. Für die Finanzwirtschaft stand ein Handelssimulator zu Verfügung, der auf Basis von echten Finanzdaten die Qualität verschiedener Strategien testen kann. Dieser wurde im Projekt weiterentwickelt. GP wurde genutzt, um auf Basis der Simulationen genaue Vorhersagen und damit verbesserte Handelsstrategien zu entwerfen. Auch für die Wasserwirtschaft wurden Prognoseverfahren mit GP entwickelt. Der Schwerpunkt lag dabei auf der Füllstandprognose für Regenüberlaufbecken. Hier konnten moderne Verfahren mit GP oder SVM klassische Methoden deutlich schlagen oder verbessern. Auch der Einsatz von Sequentieller Parameter Optimierung zeigte signifikante Verbesserungen für die Prognosegenauigkeit. Dabei war die Kombination von klassischen Methoden und GP besonders erfolgreich. GP ist nach wie vor ein sehr aktives Forschungsgebiet und erlaubt auch für die Folgezeit zahlreiche Kooperationen mit den Partnern der Fachhochschule Köln. Sowohl für technische Anwendungen als auch zur Lösung von Forschungsfragen bieten sich zahlreiche Möglichkeiten an.
The performance of optimization algorithms relies crucially on their parameterizations. Finding good parameter settings is called algorithm tuning. Using
a simple simulated annealing algorithm, we will demonstrate how optimization algorithms can be tuned using the Sequential Parameter Optimization Toolbox (SPOT). SPOT provides several tools for automated and interactive tuning. The underlying concepts of the SPOT approach are explained. This includes key techniques such as exploratory fitness landscape analysis and response surface methodology. Many examples illustrate
how SPOT can be used for understanding the performance of algorithms and gaining insight into algorithm behavior. Furthermore, we demonstrate how SPOT can be used as an optimizer and how a sophisticated ensemble approach is able to combine several meta models via stacking.
Sequential Parameter Optimization is a model-based optimization methodology, which includes several techniques for handling uncertainty. Simple approaches such as sharp- ening and more sophisticated approaches such as optimal computing budget allocation are available. For many real world engineering problems, the objective function can be evaluated at different levels of fidelity. For instance, a CFD simulation might provide a very time consuming but accurate way to estimate the quality of a solution.The same solution could be evaluated based on simplified mathematical equations, leading to a cheaper but less accurate estimate. Combining these different levels of fidelity in a model-based optimization process is referred to as multi-fidelity optimization. This chapter describes uncertainty-handling techniques for meta-model based search heuristics in combination with multi-fidelity optimization. Co-Kriging is one power- ful method to correlate multiple sets of data from different levels of fidelity. For the first time, Sequential Parameter Optimization with co-Kriging is applied to noisy test functions. This study will introduce these techniques and discuss how they can be applied to real-world examples.