Refine
Year of publication
Document Type
- Working Paper (114) (remove)
Language
- English (114) (remove)
Has Fulltext
- yes (114)
Is part of the Bibliography
- no (114) (remove)
Keywords
- Lambda-Kalkül (18)
- Formale Semantik (10)
- Operationale Semantik (8)
- Programmiersprache (7)
- lambda calculus (7)
- Nebenläufigkeit (6)
- functional programming (6)
- concurrency (5)
- pi-calculus (5)
- semantics (5)
Institute
- Informatik (114) (remove)
Classically, encoding of images by only a few, important components is done by the Principal Component Analysis (PCA). Recently, a data analysis tool called Independent Component Analysis (ICA) for the separation of independent influences in signals has found strong interest in the neural network community. This approach has also been applied to images. Whereas the approach assumes continuous source channels mixed up to the same number of channels by a mixing matrix, we assume that images are composed by only a few image primitives. This means that for images we have less sources than pixels. Additionally, in order to reduce unimportant information, we aim only for the most important source patterns with the highest occurrence probabilities or biggest information called „Principal Independent Components (PIC)“. For the example of a synthetic picture composed by characters this idea gives us the most important ones. Nevertheless, for natural images where no a-priori probabilities can be computed this does not lead to an acceptable reproduction error. Combining the traditional principal component criteria of PCA with the independence property of ICA we obtain a better encoding. It turns out that this definition of PIC implements the classical demand of Shannon’s rate distortion theory.
The efficient management of large multimedia databases requires the development of new techniques to process, characterize, and search for multimedia objects. Especially in the case of image data, the rapidly growing amount of documents prohibits a manual description of the images’ content. Instead, the automated characterization is highly desirable to support annotation and retrieval of digital images. However, this is a very complex and still unsolved task. To contribute to a solution of this problem, we have developed a mechanism for recognizing objects in images based on the query by example paradigm. Therefore, the most salient image features of an example image representing the searched object are extracted to obtain a scale-invariant object model. The use of this model provides an efficient and robust strategy for recognizing objects in images independently of their size. Further applications of the mechanism are classical recognition tasks such as scene decomposition or object tracking in video sequences.
Performance and storage requirements of topology-conserving maps for robot manipulator control
(1989)
A new programming paradigm for the control of a robot manipulator by learning the mapping between the Cartesian space and the joint space (inverse Kinematic) is discussed. It is based on a Neural Network model of optimal mapping between two high-dimensional spaces by Kohonen. This paper describes the approach and presents the optimal mapping, based on the principle of maximal information gain. It is shown that Kohonens mapping in the 2-dimensional case is optimal in this sense. Furthermore, the principal control error made by the learned mapping is evaluated for the example of the commonly used PUMA robot, the trade-off between storage resources and positional error is discussed and an optimal position encoding resolution is proposed.
It is well known that artificial neural nets can be used as approximators of any continous functions to any desired degree. Nevertheless, for a given application and a given network architecture the non-trivial task rests to determine the necessary number of neurons and the necessary accuracy (number of bits) per weight for a satisfactory operation. In this paper the problem is treated by an information theoretic approach. The values for the weights and thresholds in the approximator network are determined analytically. Furthermore, the accuracy of the weights and the number of neurons are seen as general system parameters which determine the the maximal output information (i.e. the approximation error) by the absolute amount and the relative distribution of information contained in the network. A new principle of optimal information distribution is proposed and the conditions for the optimal system parameters are derived. For the simple, instructive example of a linear approximation of a non-linear, quadratic function, the principle of optimal information distribution gives the the optimal system parameters, i.e. the number of neurons and the different resolutions of the variables.
The selection of features for classification, clustering and approximation is an important task in pattern recognition, data mining and soft computing. For real-valued features, this contribution shows how feature selection for a high number of features can be implemented using mutual in-formation. Especially, the common problem for mutual information computation of computing joint probabilities for many dimensions using only a few samples is treated by using the Rènyi mutual information of order two as computational base. For this, the Grassberger-Takens corre-lation integral is used which was developed for estimating probability densities in chaos theory. Additionally, an adaptive procedure for computing the hypercube size is introduced and for real world applications, the treatment of missing values is included. The computation procedure is accelerated by exploiting the ranking of the set of real feature values especially for the example of time series. As example, a small blackbox-glassbox example shows how the relevant features and their time lags are determined in the time series even if the input feature time series determine nonlinearly the output. A more realistic example from chemical industry shows that this enables a better ap-proximation of the input-output mapping than the best neural network approach developed for an international contest. By the computationally efficient implementation, mutual information becomes an attractive tool for feature selection even for a high number of real-valued features.
We propose a variation of online paging in two-level memory systems where pages in the fast cache get modified and therefore have to be explicitly written back to the slow memory upon evictions. For increased performance, up to alpha arbitrary pages can be moved from the cache to the slow memory within a single joint eviction, whereas fetching pages from the slow memory is still performed on a one-by-one basis. The main objective in this new alpha-paging scenario is to bound the number of evictions. After providing experimental evidence that alpha-paging can adequately model flash-memory devices in the context of translation layers we turn to the theoretical connections between alpha-paging and standard paging. We give lower bounds for deterministic and randomized alpha-paging algorithms. For deterministic algorithms, we show that an adaptation of LRU is strongly competitive, while for the randomized case we show that by adapting the classical Mark algorithm we get an algorithm with a competitive ratio larger than the lower bound by a multiplicative factor of approximately 1.7.
We consider matching, rewriting, critical pairs and the Knuth-Bendix confluence test on rewrite rules in a nominal setting extended by atom-variables. Computing critical pairs is done using nominal unification, and rewriting using nominal matching. We utilise atom-variables to formulate rewrite rules, which is an improvement over previous approaches, using usual nominal unification, nominal matching and nominal equivalence of expressions coupled with a freshness constraint. We determine the complexity of several problems in a quantified freshness logic. In particular we show that nominal matching is Πp2-complete. We prove that the adapted Knuth-Bendix confluence test is applicable to a nominal rewrite system with atom-variabes and thus, that there is a decidable test whether confluence of the ground instance of the abstract rewrite system holds. We apply the nominal Knuth Bendix confluence criterion to the theory of monads, and compute a convergent nominal rewrite system modulo alpha-equivalence.
We study the effect of randomness in the adversarial queueing model. All proofs of instability for deterministic queueing strategies exploit a finespun strategy of insertions by an adversary. If the local queueing decisions in the network are subject to randomness, it is far from obvious, that an adversary can still trick the network into instability. We show that uniform queueing is unstable even against an oblivious adversary. Consequently, randomizing the queueing decisions made to operate a network is not in itself a suitable fix for poor network performances due to packet pileups.
It is known that deterministic finite automata (DFAs) can be algorithmically minimized, i.e., a DFA M can be converted to an equivalent DFA M' which has a minimal number of states. The minimization can be done efficiently [6]. On the other hand, it is known that unambiguous finite automata (UFAs) and nondeterministic finite automata (NFAs) can be algorithmically minimized too, but their minimization problems turn out to be NP-complete and PSPACE-complete [8]. In this paper, the time complexity of the minimization problem for two restricted types of finite automata is investigated. These automata are nearly deterministic, since they only allow a small amount of non determinism to be used. On the one hand, NFAs with a fixed finite branching are studied, i.e., the number of nondeterministic moves within every accepting computation is bounded by a fixed finite number. On the other hand, finite automata are investigated which are essentially deterministic except that there is a fixed number of different initial states which can be chosen nondeterministically. The main result is that the minimization problems for these models are computationally hard, namely NP-complete. Hence, even the slightest extension of the deterministic model towards a nondeterministic one, e.g., allowing at most one nondeterministic move in every accepting computation or allowing two initial states instead of one, results in computationally intractable minimization problems.