Anca Maria Ivanescu

Modeling image similarity by Gaussian mixture models and the Signature Quadratic Form Distance

Modeling image similarity for browsing and searching in voluminous image databases is a challenging task of nearly all content-based image retrieval systems. One promising way of defining image similarity consists in applying distance-based similarity measures on compact image representations. Beyond feature histograms and feature signatures, more general feature representations are mixture models of which the Gaussian mixture model is the most prominent one. This feature representation can be compared by employing approximations of the Kullback-Leibler Divergence. Although several of those approximations have been successfully applied to model image similarity, their applicability to mixture models based on high-dimensional feature descriptors is questionable. In this paper, we thus introduce the Signature Quadratic Form Distance to measure the distance between two Gaussian mixture models of high-dimensional feature descriptors. We show the analytical computation of the proposed Gaussian Quadratic Form Distance and evaluate its retrieval performance by making use of different benchmark image databases.

Efficient processing of multiple DTW queries in time series databases

Dynamic Time Warping (DTW) is a widely used distance measure for time series that has been successfully used in science and many other application domains. As DTW is computationally expensive, there is a strong need for efficient query processing algorithms. Such algorithms exist for single queries. In many of todays applications, however, large numbers of queries arise at any given time. Existing DTW techniques do not process multiple DTW queries simultaneously, a serious limitation which slows down overall processing. In this paper, we propose an efficient processing approach for multiple DTW queries. We base our approach on the observation that algorithms in areas such as data mining and interactive visualization incur many queries that share certain characteristics. Our solution exploits these shared characteristics by pruning database time series with respect to sets of queries, and we prove a lower-bounding property that guarantees no false dismissals. Our technique can be flexibly combined with existing DTW lower bounds or other single DTW query speed-up techniques for further runtime reduction. Our thorough experimental evaluation demonstrates substantial performance gains for multiple DTW queries.

Applying similarity search for the investigation of the fuel injection process

We introduce a distance-based similarity model with application to the optimization of the fuel injection process. Our model allows for an automatic evaluation of huge and complex amount of experimental data originated from optical measurement techniques analyzing the fuel injection process. The goal is to enable researchers to get deeper insight into this process based on an automatically driven analysis.

Analysis of length and orientation of microtubules in wide-field fluorescence microscopy

In this paper we present a novel approach for the analysis of microtubules in wide-field fluorescence microscopy. Microtubules are flexible elongated structures and part of the cytoskeleton, a cytoplasmic scaffolding responsible for cell stability and motility. The method allows for precise measurements of microtubule length and orientation under different conditions despite a high variability of image data and in the presence of artefacts. Application of the proposed method to demonstrate the effect of the protein GAR22 on the rate of polymerisation of microtubules illustrates the potential of our approach.

A hybrid control approach for low temperature combustion engine control

In this paper, a hybrid control approach for low temperature combustion engines is presented. The identification as well as the controller design are demonstrated. In order to identify piecewise affine models, we propose to use correlation clustering algorithms, which are developed and used in the field of data mining. We outline the identification of the low temperature combustion engine from measurement data based on correlation clustering. The output of the identified model reproduces the measurement data of the engine very well. Based on this piecewise affine model of the process, a hybrid model predictive controller is considered. It can be shown that the hybrid controller is able to produce better control results than a model predictive controller using a single linear model. The main advantage is that the hybrid controller is able to manage the system characteristics of different operating points for each prediction step.

ClasSi: measuring ranking quality in the presence of object classes with similarity information

The quality of rankings can be evaluated by computing their correlation to an optimal ranking. State of the art ranking correlation coefficients like Kendalls τ and Spearmans ρ do not allow for the user to specify similarities between differing object classes and thus treat the transposition of objects from similar classes the same way as that of objects from dissimilar classes. We propose ClasSi, a new ranking correlation coefficient which deals with class label rankings and employs a class distance function to model the similarities between the classes. We also introduce a graphical representation of ClasSi akin to the ROCcurve which describes how the correlation evolves throughout the ranking.

The ClasSi coefficient for the evaluation of ranking quality in the presence of class similarities

Evaluation measures play an important role in the design of new approaches, and often quality is measured by assessing the relevance of the obtained result set. While many evaluation measures based on precision/recall are based on a binary relevance model, ranking correlation coefficients are better suited for multi-class problems. State-of-the-art ranking correlation coefficients like Kendall’s τ and Spearman’s ρ do not allow the user to specify similarities between differing object classes and thus treat the transposition of objects from similar classes the same way as that of objects from dissimilar classes. We propose ClasSi, a new ranking correlation coefficient which deals with class label rankings and employs a class distance function to model the similarities between the classes. We also introduce a graphical representation of ClasSi which describes how the correlation evolves throughout the ranking.

Employing the Principal Hessian Direction for Building Hinging Hyperplane Models

In this paper we address the problem of identifying a continuous nonlinear model from a set of discrete observations. The goal is to build a compact and accurate model of an underlying process, which is interpretable by the user, and can be also used for prediction purposes. Hinging hyper plane models are well suited to represent continuous piecewise linear models, but the hinge finding algorithm is guaranteed to converge only in local optima, and hence heavily depends on the initialization. We employ the principal Hessian direction to incorporate the geometrical information of the regression surface in the hinge finding process and can thus avoid the several random initializations proposed in the literature.

Employing correlation clustering for the identification of piecewise affine models

To analyze and control a system, a model is build which describes the relationship between the inputs and the corresponding outputs. While simple systems can be described by a single linear model, more complex systems can be approximated through an assembly of linear submodels. Such piecewise affine (PWA) models consists of several convex regions and linear submodels describing the input output relationship for each such region. The more regions are considered in the PWA model, the more accurate it describes the system. Still, in real world applications, simple models are necessary for performance reasons, hence a trade-off has to be made between the model complexity and its accuracy. In this paper we discuss the employment of correlation clustering algorithms for a robust identification of PWA models with reduced complexity.

Hinging hyperplane models for multiple predicted variables

Model-based learning for predicting continuous values involves building an explicit generalization of the training data. Simple linear regression and piecewise linear regression techniques are well suited for this task, because, unlike neural networks, they yield an interpretable model. The hinging hyperplane approach is a nonlinear learning technique which computes a continuous model. It consists of linear submodels over individual partitions in the regressor space. However, it is only designed for one predicted variable. In the case of r predicted variables the number of partitions grows quickly with r and the result is no longer being compact or interpretable. We propose a generalization of the hinging hyperplane approach for several predicted variables. The algorithm considers all predicted variables simultaneously. It enforces common hinges, while at the same time restoring the continuity of the resulting functions. The model complexity no longer depends on the number of predicted variables, remaining compact and interpretable.