Content- and semantic-based interactive mining systems describe remote sensing images by means of relevant features. Region-based retrieval systems have been proposed to capture the local properties of an image. Existing systems use computationally extensive methods to extract primitive features based on color, texture (spatial gray level dependency - SGLD matrices), and shape from the segmented homogenous region. The use of wavelet transform techniques has recently gained momentum in multimedia image archives to expedite the retrieval process. However, the current semantic-enabled framework for the geospatial data uses computationally extensive methods for feature extraction and image segmentation. Hence, this dissertation presents the use of a wavelet-based feature extraction in a semantics-enabled framework to expedite the knowledge discovery in geospatial data archives. Geospatial data has different characteristics than multimedia images and poses more challenges. The experimental assumptions, such as the selection of the wavelet decomposition level and mother wavelet used for multimedia data archives, might not prove to be efficient for the retrieval of geospatial data. Discrete wavelet transforms (DWT) introduce aliasing effects due to subband decimation at a certain decomposition level. This dissertation addresses the issue of selecting a suitable wavelet decomposition level, and a systematic selection process is developed for image segmentation. To validate the applicability of this method, a synthetic image is generated to assess the performance qualitatively and quantitatively. In addition, results for a Landsat7 ETM+ imagery archive are illustrated, and the F-measure is used to assess the feasibility of this method for retrieval of different classes. This dissertation also introduces a new feature set obtained by coalescing wavelet and independent component analysis for image information mining. Feature-level fusion is performed to include the missing high detail information from the panchromatic image. Results show that the presented feature set is computationally less expensive and more efficient in capturing the spectral and spatial texture information when compared to traditional approaches. After extensive experimentation with different types of mother wavelets, it can be concluded that reverse Biorthogonal wavelets of shorter length and the simple Haar filter provided better results for the image information mining from the database used in this study.

Bayesian learning methods are the basis of many powerful analysis techniques in neuroimaging, permitting probabilistic inference on hierarchical, generative models of data. This thesis primarily develops Bayesian analysis techniques for magnetic resonance imaging (MRI), which is a noninvasive neuroimaging tool for probing function, perfusion, and structure in the human brain. The first part of this work fits nonlinear biophysical models to multimodal functional MRI data within a variational Bayes framework. Simultaneously-acquired multimodal data contains mixtures of different signals and therefore may have common noise sources, and a method for automatically modelling this correlation is developed. A Gaussian process prior is also used to allow spatial regularization while simultaneously applying informative priors on model parameters, restricting biophysically-interpretable parameters to reasonable values. The second part introduces a novel data fusion framework for multivariate data analysis which finds a joint decomposition of data across several modalities using a shared loading matrix. Each modality has its own generative model, including separate spatial maps, noise models and sparsity priors. This flexible approach can perform supervised learning by using target variables as a modality. By inferring the data decomposition and multivariate decoding simultaneously, the decoding targets indirectly influence the component shapes and help to preserve useful components. The same framework is used for unsupervised learning by placing independent component analysis (ICA) priors on the spatial maps. Linked ICA is a novel approach developed to jointly decompose multimodal data, and is applied to combined structural and diffusion images across groups of subjects. This allows some of the benefits of tensor ICA and spatially-concatenated ICA to be combined, and allows model comparison between different configurations. This joint decomposition framework is particularly flexible because of its separate generative models for each modality and could potentially improve modelling of functional MRI, magnetoencephalography, and other functional neuroimaging modalities.

L'analyse de données d'expression montre qu'il est avantageux d'analyser les processus biologiques en termes de modules plutôt que simplement considérer les gènes un par un. Dans ce projet nous avons conduit une analyse non supervisée des données d'expression de gènes de plusieurs cohortes de tumeurs urothéliales en appliquant la méthode d'Analyse en Composantes Indépendantes. Plusieurs études ont démontré les meilleures performances de l'ACI par rapport à l'ACP et les méthodes de clustering, pour obtenir une décomposition plus réaliste des données d'expression en patterns d'expression pertinents et associés avec le phénotype d'intérêt.Les tumeurs urothéliales apparaissent et évoluent selon deux voies distinctes dont la probabilité de progression en cancer musculo-invasif diffère radicalement. Excepté la mutation de FGFR3 dans le groupe le moins agressif, les processus moléculaires sous-jacents n'ont pas été complètement identifiés. Le principal objectif de cette thèse était dédié aux interprétations biologiques des différentes composantes indépendantes pour aider à confirmer et étendre la liste des processus biologiques connus pour être impliqués dans le cancer de vessie.Chaque composante indépendante est caractérisée par une liste de projections de gènes et de contributions pondérées d'échantillons tumoraux . En combinant expertise biologique et comparaison des listes de gènes à des voies existantes et en étudiant conjointement l'association des composantes aux annotations cliniques et moléculaires, nous avons pu différencier les CIscausées par des facteurs techniques, tels que le prélèvement chirurgical de celles ayant des interprétations biologiques pertinentes. De plus, parmi les signaux pertinents biologiquement, cette analyse nous a permis de différencier les signaux provenant du stroma, comme la réponse immunitaire médiée par les lymphocytesB&T, de ceux produits par les tumeurs elles-mêmes, comme les signaux reliés à la prolifération ou à la différenciation. La classification des tumeurs selon leurs contributions à certaines CIs a pu être étroitement associée à des annotations anatomo-cliniques, et a mis en évidence de nouveaux sous-types de tumeur spotentiels, qui suggèrent l'existence de voies de progression supplémentaires dans le cancer de vessie. De façon similaire, l'étude des contributions de groupes de tumeurs basés sur des annotations cliniques ou moléculaires a montré différents niveaux de contamination par le stroma entre les tumeurs mutées et nonmutées pour FGFR3. La reproductibilité des composantes a été étudiée en utilisant des graphes de corrélation. La majeure partie des CIs interprétées a été validée sur trois jeux de données indépendants, et plusieurs d'entre elles ont aussi détectées dans un jeu de données de lignées cellulaires.Une deuxième étude sur le rétinoblastome a montré que nous pouvions tirer partie de l'ACI dans des contextes variés. Le rétinoblastome est initié par la perte des deux alléles du gène suppresseur de tumeur RB1. D'autres évènements génomiques non identifiés sont nécessaires à la progression de la maladie. Nous avons observé une association entre âge des patients et altérations génomiques. Les patients jeunes présentant moins d'altérations que les patients âgés, ces derniers présentant des gains du 1q et des pertes du 16q. Cette séparation des tumeurs selon l'âge est également observée sur les données d'expression, notamment en appliquant l'ACI dont l'une des composantes discrimine les patients selon leur âge. Ces résultats suggèrent l'existence de deux voies de progression dans le rétinoblastome. L'analyse des données à haut débit fournit de nombreuses listes de gènes. Afin de les interpréter, une possibilité est d'extraire les dernières publications groupées par sujets prédéfinis (fonction, localisation,...). / Practice of gene expression data analysis shows that it is advantageous to analyze biologicalprocesses in terms of modules rather than simply consider gene one by one. In this project, we conducted anunsupervised analysis of the gene expression data of several cohorts of urothelial tumors, applying theIndependent Component Analysis method. Several studies demonstrated the outperformance of ICA overPCA and clustering-based methods in obtaining a more realistic decomposition of the expression data intoconsistent patterns of coexpressed genes associated with the studied phenotypes[1, 2, 3, 4].Urothelial tumors arise and evolve through two distinct pathways which radically differ on their probabilityof progression to muscle invasion. Except the mutation of FGFR3 in the less aggressive group, theunderlying molecular processes have not been completely identified. The first and main objective of the PhDthesis was dedicated to the biological interpretation of the different independent components to help toconfirm and extend the list of biological processes known to be involved in bladder cancer.Each independent component (IC) is characterized by a list of gene projections on the one hand and weightedcontributions of tumor samples on the other hand. By combining biological expertise and comparison of theassociated list of genes to known pathways, and jointly studying the association of the components tomolecular and clinical annotations, we have been able to differentiate components that were caused bytechnical factors, such as surgical sampling, from those having consistent biological interpretationin terms of tumor biology. Moreover, among the biologically meaningful signals, this analysis allowed us todifferentiate the signals from stroma of the tumor, like immune response mediated by B- and T-lymphocytes,from the signals produced by the tumors themselves, like signals related to proliferation, or differentiation.The clustering of the tumor samples according to their contributions on some ICs can be closely associated toanatomo-clinical annotations, and highlighted new potential subtypes of tumors which suggest existence ofadditional pathways of bladder cancer progression. Similarly, the study of the contributions of preestablishedgroups of tumors based on clinical or molecular criteria showed different levels of stromacontamination between FGFR3 non-mutated and mutated tumors. The reproducibility of the components wasinvestigated using correlation graphs. The major part of the interpreted ICs was validated on threeindependent bladder cancer datasets, and several of them were also identified in an urothelial cancer celllines data set.A second study about retinoblastoma gave us the occasion to show that we can take advantage ofICA in various contexts. Retinoblastoma is initiated by the loss of both alleles of the RB1 tumor suppressorgene. Although necessary for initiation, other genomic events, that remain to be identified, are needed for theprogression of the disease [5]. We observed, as it was previously described [6], an association between age ofthe patients and levels of genomic aberrations, the younger patients having fewer alterations than the olderpatients, which generally present gain of 1q and loss of 16q. We showed that this tendency of the tumors tobe clustered into two groups of age can also be observed on the expression data by applying ICA whose oneof the component was highly correlated to the age of the patients. These results suggest the existence of twopathways of progression in retinoblastoma.The analysis of high throughput data provides many lists of genes. To interpret them, a possibility isto study the latest related publications grouped by pre-defined group of topics (function, cellular location...).To that aim, in a third study, we introduced a web-based Java application tool named GeneValorization whichgives a clear and handful overview of the bibliography corresponding to one particular gene list [7].

This thesis has evaluated the use of Independent Component Analysis (ICA) on Surface Electromyography (sEMG), focusing on the biosignal applications. This research has identified and addressed the following four issues related to the use of ICA for biosignals: • The iterative nature of ICA • The order and magnitude ambiguity problems of ICA • Estimation of number of sources based on dependency and independency nature of the signals • Source separation for non-quadratic ICA (undercomplete and overcomplete) This research first establishes the applicability of ICA for sEMG and also identifies the shortcomings related to order and magnitude ambiguity. It has then developed, a mitigation strategy for these issues by using a single unmixing matrix and neural network weight matrix corresponding to the specific user. The research reports experimental verification of the technique and also the investigation of the impact of inter-subject and inter-experimental variations. The results demonstrate that while using sEMG without separation gives only 60% accuracy, and sEMG separated using traditional ICA gives an accuracy of 65%, this approach gives an accuracy of 99% for the same experimental data. Besides the marked improvement in accuracy, the other advantages of such a system are that it is suitable for real time operations and is easy to train by a lay user. The second part of this thesis reports research conducted to evaluate the use of ICA for the separation of bioelectric signals when the number of active sources may not be known. The work proposes the use of value of the determinant of the Global matrix generated using sparse sub band ICA for identifying the number of active sources. The results indicate that the technique is successful in identifying the number of active muscles for complex hand gestures. The results support the applications such as human computer interface. This thesis has also developed a method of determining the number of independent sources in a given mixture and has also demonstrated that using this information, it is possible to separate the signals in an undercomplete situation and reduce the redundancy in the data using standard ICA methods. The experimental verification has demonstrated that the quality of separation using this method is better than other techniques such as Principal Component Analysis (PCA) and selective PCA. This has number of applications such as audio separation and sensor networks.

Stimulus-driven changes in the gain of sensory neurons are well-documented, but relatively little is known about whether analogous gain-control can also be effected in a top-down manner. A recent psychophysical study demonstrated that sensitivity to luminance contrast can be modulated by a priori knowledge (de la Rosa et al., in press). In the present study, event-related potentials were used to resolve the stages of information processing that facilitate such knowledge-driven adjustments. Groupwise independent component analysis identified two robust spatiotemporal patterns of endogenous brain activity that captured experimental effects. The first pattern was associated with obligatory processing of contextual information, while the second pattern was associated with selective initiation of contrast gain adjustment. These data suggest that knowledge-driven contrast gain control is mediated by multiple independent electrogenic sources.

Stimulus-driven changes in the gain of sensory neurons are well-documented, but relatively little is known about whether analogous gain-control can also be effected in a top-down manner. A recent psychophysical study demonstrated that sensitivity to luminance contrast can be modulated by a priori knowledge (de la Rosa et al., in press). In the present study, event-related potentials were used to resolve the stages of information processing that facilitate such knowledge-driven adjustments. Groupwise independent component analysis identified two robust spatiotemporal patterns of endogenous brain activity that captured experimental effects. The first pattern was associated with obligatory processing of contextual information, while the second pattern was associated with selective initiation of contrast gain adjustment. These data suggest that knowledge-driven contrast gain control is mediated by multiple independent electrogenic sources.

Music recordings contain the mixed contribution of multiple overlapping instruments. In order to better understand the music, it would be beneficial to understand each instrument independently. This thesis focuses on separating the individual instrument recordings within a song. In particular, we propose novel algorithms for separating instrument recordings given only their mixture. When the number of source signals does not exceed the number of mixture signals, we focus on a subclass of source separation algorithms based on joint diagonalization. Each approach leverages a different form of source structure. We introduce repetitive structure as an alternative that leverages unique repetition patterns in music and compare its performance against the other techniques. When the number of source signals exceeds the number of mixtures (i.e. the underdetermined problem), we focus on spectrogram factorization techniques for source separation. We extend single-channel techniques to utilize the additional spatial information in multichannel recordings, and use phase information to improve the estimation of the underlying components.

No description available.

Σκοπός της εργασίας είναι η ανάπτυξη μεθόδου εντοπισμού εγκεφαλικών ρυθμών στο χρόνο χρησιμοποιώντας περιορισμούς που στηρίζονται στα νευροφυσιολογικά χαρακτηριστικά του κάθε υθμού τόσο στο πεδίο του χώρου (spatial constraints) όσο και στο πεδίο της συχνότητας (frequency constraints). Η μέθοδος στηρίζεται στην τεχνική ανάλυσης σε ανεξάρτητες συνιστώσες (ICA) και δεν απαιτεί πολυκάναλες καταγραφές (MEG). Πιθανές εφαρμογές περιλαμβάνουν τον εντοπισμό α-ρυθμού, επιληπτικών κρίσεων, μ-ρυθμού και ρυθμών κυρίαρχων στα στάδια του ύπνου. Η προτεινόμενη μέθοδος μπορεί να χρησιμοποιηθεί για την ανάλυση καταγραφών ΗΕΓ τόσο σε πραγματικό χρόνο (online) όσο και σε προαποθηκευμένα δεδομένα (offline). / The goal of the present thesis is the development of a method for temporal detection of electrophysiological brain rhythms, using constraints based on neurophysiological, spatial and frequency characteristics of every rhythm. The method is based on Independent Component Analysis (ICA) and does not require multichannel recordings (MEG). Possible applications include temporal detection of α-rhythm, μ-rhythm and sleep dominant rhythms. The proposed method can be used in both online and offline EEG analysis.

This dissertation explored the way in which bumblebees' visual system helps them discover their first flower. Previous studies found bees have unlearned preferences for parts of a flower, such as its colour and shape. The first study pitted two variables against each other: pattern type: sunburst or bull's eye, versus the location of the pattern: shapes appeared peripherally or centrally. We observed free-flying bees in a flight cage using Radio-Frequency Identification (RFID) tracking. The results show two distinct behavioural preferences: Pattern type predicts landing: bees prefer radial over concentric patterns, regardless of whether the radial pattern is on the perimeter or near the centre of the flower. Pattern location predicts exploration: bees were more likely to explore the inside of artificial flowers if the shapes were displayed near the centre of the flower, regardless of whether the pattern was radial or concentric. As part of the second component, we implemented a mathematical model aimed at explaining how bees come to prefer radial patterns, leafy backgrounds and symmetry. The model was based on unsupervised neural networks used to describe cognitive mechanisms. The results captured with the results of multiple behavioural experiments. The model suggests that bees choose computationally "cheaper" stimuli, those that contain less information. The third study tested the computational load hypothesis generated by the artificial neural networks. Visual properties of symmetry, and spatial frequency were tested. Studying free-flying bees in a flight cage using motion-sensitive video recordings, we found that bees preferred 4-axis symmetrical patterns in both low and high frequency displays.