Lernen mit einer Hybride aus Lernendem Klassifizierendem System und Selbstorganisierender Karte

Im Forschungsgebiet der Künstlichen Intelligenz, insbesondere im Bereich des maschinellen Lernens, hat sich eine ganze Reihe von Verfahren etabliert, die von biologischen Vorbildern inspiriert sind. Die prominentesten Vertreter derartiger Verfahren sind zum einen Evolutionäre Algorithmen, zum anderen Künstliche Neuronale Netze. Die vorliegende Arbeit befasst sich mit der Entwicklung eines Systems zum maschinellen Lernen, das Charakteristika beider Paradigmen in sich vereint: Das Hybride Lernende Klassifizierende System (HCS) wird basierend auf dem reellwertig kodierten eXtended Learning Classifier System (XCS), das als Lernmechanismus einen Genetischen Algorithmus enthält, und dem Wachsenden Neuralen Gas (GNG) entwickelt. Wie das XCS evolviert auch das HCS mit Hilfe eines Genetischen Algorithmus eine Population von Klassifizierern - das sind Regeln der Form [WENN Bedingung DANN Aktion], wobei die Bedingung angibt, in welchem Bereich des Zustandsraumes eines Lernproblems ein Klassifizierer anwendbar ist. Beim XCS spezifiziert die Bedingung in der Regel einen achsenparallelen Hyperquader, was oftmals keine angemessene Unterteilung des Zustandsraumes erlaubt. Beim HCS hingegen werden die Bedingungen der Klassifizierer durch Gewichtsvektoren beschrieben, wie die Neuronen des GNG sie besitzen. Jeder Klassifizierer ist anwendbar in seiner Zelle der durch die Population des HCS induzierten Voronoizerlegung des Zustandsraumes, dieser kann also flexibler unterteilt werden als beim XCS. Die Verwendung von Gewichtsvektoren ermöglicht ferner, einen vom Neuronenadaptationsverfahren des GNG abgeleiteten Mechanismus als zweites Lernverfahren neben dem Genetischen Algorithmus einzusetzen. Während das Lernen beim XCS rein evolutionär erfolgt, also nur durch Erzeugen neuer Klassifizierer, ermöglicht dies dem HCS, bereits vorhandene Klassifizierer anzupassen und zu verbessern. Zur Evaluation des HCS werden mit diesem verschiedene Lern-Experimente durchgeführt. Die Leistungsfähigkeit des Ansatzes wird in einer Reihe von Lernproblemen aus den Bereichen der Klassifikation, der Funktionsapproximation und des Lernens von Aktionen in einer interaktiven Lernumgebung unter Beweis gestellt.

Learning Methods and Algorithms for Semantic Text Classification across Multiple Domains

Information is nowadays a key resource: machine learning and data mining techniques have been developed to extract high-level information from great amounts of data. As most data comes in form of unstructured text in natural languages, research on text mining is currently very active and dealing with practical problems. Among these, text categorization deals with the automatic organization of large quantities of documents in priorly defined taxonomies of topic categories, possibly arranged in large hierarchies. In commonly proposed machine learning approaches, classifiers are automatically trained from pre-labeled documents: they can perform very accurate classification, but often require a consistent training set and notable computational effort. Methods for cross-domain text categorization have been proposed, allowing to leverage a set of labeled documents of one domain to classify those of another one. Most methods use advanced statistical techniques, usually involving tuning of parameters. A first contribution presented here is a method based on nearest centroid classification, where profiles of categories are generated from the known domain and then iteratively adapted to the unknown one. Despite being conceptually simple and having easily tuned parameters, this method achieves state-of-the-art accuracy in most benchmark datasets with fast running times. A second, deeper contribution involves the design of a domain-independent model to distinguish the degree and type of relatedness between arbitrary documents and topics, inferred from the different types of semantic relationships between respective representative words, identified by specific search algorithms. The application of this model is tested on both flat and hierarchical text categorization, where it potentially allows the efficient addition of new categories during classification. Results show that classification accuracy still requires improvements, but models generated from one domain are shown to be effectively able to be reused in a different one.

Machine Learning applicato al Web Semantico: Statistical Relational Learning vs Tensor Factorization

Obiettivo della tesi è analizzare e testare i principali approcci di Machine Learning applicabili in contesti semantici, partendo da algoritmi di Statistical Relational Learning, quali Relational Probability Trees, Relational Bayesian Classifiers e Relational Dependency Networks, per poi passare ad approcci basati su fattorizzazione tensori, in particolare CANDECOMP/PARAFAC, Tucker e RESCAL.

A new information fusion method for land-use classification using high resolution satellite imagery

Satellite image classification involves designing and developing efficient image classifiers. With satellite image data and image analysis methods multiplying rapidly, selecting the right mix of data sources and data analysis approaches has become critical to the generation of quality land-use maps. In this study, a new postprocessing information fusion algorithm for the extraction and representation of land-use information based on high-resolution satellite imagery is presented. This approach can produce land-use maps with sharp interregional boundaries and homogeneous regions. The proposed approach is conducted in five steps. First, a GIS layer - ATKIS data - was used to generate two coarse homogeneous regions, i.e. urban and rural areas. Second, a thematic (class) map was generated by use of a hybrid spectral classifier combining Gaussian Maximum Likelihood algorithm (GML) and ISODATA classifier. Third, a probabilistic relaxation algorithm was performed on the thematic map, resulting in a smoothed thematic map. Fourth, edge detection and edge thinning techniques were used to generate a contour map with pixel-width interclass boundaries. Fifth, the contour map was superimposed on the thematic map by use of a region-growing algorithm with the contour map and the smoothed thematic map as two constraints. For the operation of the proposed method, a software package is developed using programming language C. This software package comprises the GML algorithm, a probabilistic relaxation algorithm, TBL edge detector, an edge thresholding algorithm, a fast parallel thinning algorithm, and a region-growing information fusion algorithm. The county of Landau of the State Rheinland-Pfalz, Germany was selected as a test site. The high-resolution IRS-1C imagery was used as the principal input data.

Meta-analyses qualify metzincins and related genes as acute rejection markers in renal transplant patients

Definition of acute renal allograft rejection (AR) markers remains clinically relevant. Features of T-cell-mediated AR are tubulointerstitial and vascular inflammation associated with excessive extracellular matrix (ECM) remodeling, regulated by metzincins, including matrix metalloproteases (MMP). Our study focused on expression of metzincins (METS), and metzincins and related genes (MARGS) in renal allograft biopsies using four independent microarray data sets. Our own cases included normal histology (N, n = 20), borderline changes (BL, n = 4), AR (n = 10) and AR + IF/TA (n = 7). MARGS enriched in all data sets were further examined on mRNA and/or protein level in additional patients. METS and MARGS differentiated AR from BL, AR + IF/TA and N in a principal component analysis. Their expression changes correlated to Banff t- and i-scores. Two AR classifiers, based on METS (including MMP7, TIMP1), or on MARGS were established in our own and validated in the three additional data sets. Thirteen MARGS were significantly enriched in AR patients of all data sets comprising MMP7, -9, TIMP1, -2, thrombospondin2 (THBS2) and fibrillin1. RT-PCR using microdissected glomeruli/tubuli confirmed MMP7, -9 and THBS2 microarray results; immunohistochemistry showed augmentation of MMP2, -9 and TIMP1 in AR. TIMP1 and THBS2 were enriched in AR patient serum. Therefore, differentially expressed METS and MARGS especially TIMP1, MMP7/-9 represent potential molecular AR markers.

Time-lapse microscopy and classification of 2D human mesenchymal stem cells based on cell shape picks up myogenic from osteogenic and adipogenic differentiation

Current methods to characterize mesenchymal stem cells (MSCs) are limited to CD marker expression, plastic adherence and their ability to differentiate into adipogenic, osteogenic and chondrogenic precursors. It seems evident that stem cells undergoing differentiation should differ in many aspects, such as morphology and possibly also behaviour; however, such a correlation has not yet been exploited for fate prediction of MSCs. Primary human MSCs from bone marrow were expanded and pelleted to form high-density cultures and were then randomly divided into four groups to differentiate into adipogenic, osteogenic chondrogenic and myogenic progenitor cells. The cells were expanded as heterogeneous and tracked with time-lapse microscopy to record cell shape, using phase-contrast microscopy. The cells were segmented using a custom-made image-processing pipeline. Seven morphological features were extracted for each of the segmented cells. Statistical analysis was performed on the seven-dimensional feature vectors, using a tree-like classification method. Differentiation of cells was monitored with key marker genes and histology. Cells in differentiation media were expressing the key genes for each of the three pathways after 21 days, i.e. adipogenic, osteogenic and chondrogenic, which was also confirmed by histological staining. Time-lapse microscopy data were obtained and contained new evidence that two cell shape features, eccentricity and filopodia (= 'fingers') are highly informative to classify myogenic differentiation from all others. However, no robust classifiers could be identified for the other cell differentiation paths. The results suggest that non-invasive automated time-lapse microscopy could potentially be used to predict the stem cell fate of hMSCs for clinical application, based on morphology for earlier time-points. The classification is challenged by cell density, proliferation and possible unknown donor-specific factors, which affect the performance of morphology-based approaches. Copyright © 2012 John Wiley & Sons, Ltd.

Tumor classification of six common cancer types based on proteomic profiling by MALDI imaging

In clinical diagnostics, it is of outmost importance to correctly identify the source of a metastatic tumor, especially if no apparent primary tumor is present. Tissue-based proteomics might allow correct tumor classification. As a result, we performed MALDI imaging to generate proteomic signatures for different tumors. These signatures were used to classify common cancer types. At first, a cohort comprised of tissue samples from six adenocarcinoma entities located at different organ sites (esophagus, breast, colon, liver, stomach, thyroid gland, n = 171) was classified using two algorithms for a training and test set. For the test set, Support Vector Machine and Random Forest yielded overall accuracies of 82.74 and 81.18%, respectively. Then, colon cancer liver metastasis samples (n = 19) were introduced into the classification. The liver metastasis samples could be discriminated with high accuracy from primary tumors of colon cancer and hepatocellular carcinoma. Additionally, colon cancer liver metastasis samples could be successfully classified by using colon cancer primary tumor samples for the training of the classifier. These findings demonstrate that MALDI imaging-derived proteomic classifiers can discriminate between different tumor types at different organ sites and in the same site.

A Data-Analytic Strategy for Protein-Biomarker Discovery: Profiling of High-Dimensional Proteomic Data for Cancer Detection

With recent advances in mass spectrometry techniques, it is now possible to investigate proteins over a wide range of molecular weights in small biological specimens. This advance has generated data-analytic challenges in proteomics, similar to those created by microarray technologies in genetics, namely, discovery of "signature" protein profiles specific to each pathologic state (e.g., normal vs. cancer) or differential profiles between experimental conditions (e.g., treated by a drug of interest vs. untreated) from high-dimensional data. We propose a data analytic strategy for discovering protein biomarkers based on such high-dimensional mass-spectrometry data. A real biomarker-discovery project on prostate cancer is taken as a concrete example throughout the paper: the project aims to identify proteins in serum that distinguish cancer, benign hyperplasia, and normal states of prostate using the Surface Enhanced Laser Desorption/Ionization (SELDI) technology, a recently developed mass spectrometry technique. Our data analytic strategy takes properties of the SELDI mass-spectrometer into account: the SELDI output of a specimen contains about 48,000 (x, y) points where x is the protein mass divided by the number of charges introduced by ionization and y is the protein intensity of the corresponding mass per charge value, x, in that specimen. Given high coefficients of variation and other characteristics of protein intensity measures (y values), we reduce the measures of protein intensities to a set of binary variables that indicate peaks in the y-axis direction in the nearest neighborhoods of each mass per charge point in the x-axis direction. We then account for a shifting (measurement error) problem of the x-axis in SELDI output. After these pre-analysis processing of data, we combine the binary predictors to generate classification rules for cancer, benign hyperplasia, and normal states of prostate. Our approach is to apply the boosting algorithm to select binary predictors and construct a summary classifier. We empirically evaluate sensitivity and specificity of the resulting summary classifiers with a test dataset that is independent from the training dataset used to construct the summary classifiers. The proposed method performed nearly perfectly in distinguishing cancer and benign hyperplasia from normal. In the classification of cancer vs. benign hyperplasia, however, an appreciable proportion of the benign specimens were classified incorrectly as cancer. We discuss practical issues associated with our proposed approach to the analysis of SELDI output and its application in cancer biomarker discovery.

Classification Using Generalized Partial Least Squares

The advances in computational biology have made simultaneous monitoring of thousands of features possible. The high throughput technologies not only bring about a much richer information context in which to study various aspects of gene functions but they also present challenge of analyzing data with large number of covariates and few samples. As an integral part of machine learning, classification of samples into two or more categories is almost always of interest to scientists. In this paper, we address the question of classification in this setting by extending partial least squares (PLS), a popular dimension reduction tool in chemometrics, in the context of generalized linear regression based on a previous approach, Iteratively ReWeighted Partial Least Squares, i.e. IRWPLS (Marx, 1996). We compare our results with two-stage PLS (Nguyen and Rocke, 2002A; Nguyen and Rocke, 2002B) and other classifiers. We show that by phrasing the problem in a generalized linear model setting and by applying bias correction to the likelihood to avoid (quasi)separation, we often get lower classification error rates.

Land use/cover change using Remote Sensing and Geographic Information Systems : Pic Macaya National Park, Haiti

A post classification change detection technique based on a hybrid classification approach (unsupervised and supervised) was applied to Landsat Thematic Mapper (TM), Landsat Enhanced Thematic Plus (ETM+), and ASTER images acquired in 1987, 2000 and 2004 respectively to map land use/cover changes in the Pic Macaya National Park in the southern region of Haiti. Each image was classified individually into six land use/cover classes: built-up, agriculture, herbaceous, open pine forest, mixed forest, and barren land using unsupervised ISODATA and maximum likelihood supervised classifiers with the aid of field collected ground truth data collected in the field. Ground truth information, collected in the field in December 2007, and including equalized stratified random points which were visual interpreted were used to assess the accuracy of the classification results. The overall accuracy of the land classification for each image was respectively: 1987 (82%), 2000 (82%), 2004 (87%). A post classification change detection technique was used to produce change images for 1987 to 2000, 1987 to 2004, and 2000 to 2004. It was found that significant changes in the land use/cover occurred over the 17- year period. The results showed increases in built up (from 10% to 17%) and herbaceous (from 5% to 14%) areas between 1987 and 2004. The increase of herbaceous was mostly caused by the abandonment of exhausted agriculture lands. At the same time, open pine forest and mixed forest areas lost (75%) and (83%) of their area to other land use/cover types. Open pine forest (from 20% to 14%) and mixed forest (from18 to 12%) were transformed into agriculture area or barren land. This study illustrated the continuing deforestation, land degradation and soil erosion in the region, which in turn is leading to decrease in vegetative cover. The study also showed the importance of Remote Sensing (RS) and Geographic Information System (GIS) technologies to estimate timely changes in the land use/cover, and to evaluate their causes in order to design an ecological based management plan for the park.

A 45nm CMOS DESIGN OF NEUROMORPHIC SYSTEM

Neuromorphic computing has become an emerging field in wide range of applications. Its challenge lies in developing a brain-inspired architecture that can emulate human brain and can work for real time applications. In this report a flexible neural architecture is presented which consists of 128 X 128 SRAM crossbar memory and 128 spiking neurons. For Neuron, digital integrate and fire model is used. All components are designed in 45nm technology node. The core can be configured for certain Neuron parameters, Axon types and synapses states and are fully digitally implemented. Learning for this architecture is done offline. To train this circuit a well-known algorithm Restricted Boltzmann Machine (RBM) is used and linear classifiers are trained at the output of RBM. Finally, circuit was tested for handwritten digit recognition application. Future prospects for this architecture are also discussed.

Automatic Food Intake Assessment Using Camera Phones

Obesity is becoming an epidemic phenomenon in most developed countries. The fundamental cause of obesity and overweight is an energy imbalance between calories consumed and calories expended. It is essential to monitor everyday food intake for obesity prevention and management. Existing dietary assessment methods usually require manually recording and recall of food types and portions. Accuracy of the results largely relies on many uncertain factors such as user's memory, food knowledge, and portion estimations. As a result, the accuracy is often compromised. Accurate and convenient dietary assessment methods are still blank and needed in both population and research societies. In this thesis, an automatic food intake assessment method using cameras, inertial measurement units (IMUs) on smart phones was developed to help people foster a healthy life style. With this method, users use their smart phones before and after a meal to capture images or videos around the meal. The smart phone will recognize food items and calculate the volume of the food consumed and provide the results to users. The technical objective is to explore the feasibility of image based food recognition and image based volume estimation. This thesis comprises five publications that address four specific goals of this work: (1) to develop a prototype system with existing methods to review the literature methods, find their drawbacks and explore the feasibility to develop novel methods; (2) based on the prototype system, to investigate new food classification methods to improve the recognition accuracy to a field application level; (3) to design indexing methods for large-scale image database to facilitate the development of new food image recognition and retrieval algorithms; (4) to develop novel convenient and accurate food volume estimation methods using only smart phones with cameras and IMUs. A prototype system was implemented to review existing methods. Image feature detector and descriptor were developed and a nearest neighbor classifier were implemented to classify food items. A reedit card marker method was introduced for metric scale 3D reconstruction and volume calculation. To increase recognition accuracy, novel multi-view food recognition algorithms were developed to recognize regular shape food items. To further increase the accuracy and make the algorithm applicable to arbitrary food items, new food features, new classifiers were designed. The efficiency of the algorithm was increased by means of developing novel image indexing method in large-scale image database. Finally, the volume calculation was enhanced through reducing the marker and introducing IMUs. Sensor fusion technique to combine measurements from cameras and IMUs were explored to infer the metric scale of the 3D model as well as reduce noises from these sensors.

Haptische, visuelle und olfaktorische Sprachen: Westnilotische Wahrnehmungen

Cet article traite des expressions de la perception sensuelle de l’ouest nilotique. La première partie de l’article présente une terminologie ophrésiologique en louo et burun et démontre que cette catégorie lexicale détient aussi une catégorie grammaticale particulière. Phénomènes très rares dans les langues du monde, les termes ophrésiologiques sont seulement présentés sous forme introductive pour encourager davantage les recherches futures. La seconde partie de l’article porte sur les descriptions des couleurs utilisées pour les animaux domestiques. Quand les modalités de l’économie changent, les noms utilisés pour les couleurs des animaux peuvent aussi être employés pour d’autres concepts culturels. La troisième partie de l’article montre que les classificateurs nominaux en mabaan (burun) expriment des principes concernant le toucher en tant que structure cognitive. En conséquence, différents procédés de grammaticalisation sont assumés et corrélés avec des ponctuations dans l’histoire culturelle et mentale des ancêtres des locuteurs de l’ouest nilotique.

Automatic identification of discourse markers in dialogues: An in-depth study of like and well

The lexical items like and well can serve as discourse markers (DMs), but can also play numerous other roles, such as verb or adverb. Identifying the occurrences that function as DMs is an important step for language understanding by computers. In this study, automatic classifiers using lexical, prosodic/positional and sociolinguistic features are trained over transcribed dialogues, manually annotated with DM information. The resulting classifiers improve state-of-the-art performance of DM identification, at about 90% recall and 79% precision for like (84.5% accuracy, κ = 0.69), and 99% recall and 98% precision for well (97.5% accuracy, κ = 0.88). Automatic feature analysis shows that lexical collocations are the most reliable indicators, followed by prosodic/positional features, while sociolinguistic features are marginally useful for the identification of DM like and not useful for well. The differentiated processing of each type of DM improves classification accuracy, suggesting that these types should be treated individually.

Shallow Dialogue Processing Using Machine Learning Algorithms (or not)

This paper presents a shallow dialogue analysis model, aimed at human-human dialogues in the context of staff or business meetings. Four components of the model are defined, and several machine learning techniques are used to extract features from dialogue transcripts: maximum entropy classifiers for dialogue acts, latent semantic analysis for topic segmentation, or decision tree classifiers for discourse markers. A rule-based approach is proposed for solving cross-modal references to meeting documents. The methods are trained and evaluated thanks to a common data set and annotation format. The integration of the components into an automated shallow dialogue parser opens the way to multimodal meeting processing and retrieval applications.