Can neuroimaging be used as a support to diagnosis of borderline personality disorder? An approach based on computational neuroanatomy and machine learning

Several recent studies in literature have identified brain morphological alterations associated to Borderline Personality Disorder (BPD) patients. These findings are reported by studies based on voxel-based-morphometry analysis of structural MRI data, comparing mean gray-matter concentration between groups of BPD patients and healthy controls. On the other hand, mean differences between groups are not informative about the discriminative value of neuroimaging data to predict the group of individual subjects. In this paper, we go beyond mean differences analyses, and explore to what extent individual BPD patients can be differentiated from controls (25 subjects in each group), using a combination of automated-morphometric tools for regional cortical thickness/volumetric estimation and Support Vector Machine classifier. The approach included a feature selection step in order to identify the regions containing most discriminative information. The accuracy of this classifier was evaluated using the leave-one-subject-out procedure. The brain regions indicated as containing relevant information to discriminate groups were the orbitofrontal, rostral anterior cingulate, posterior cingulate, middle temporal cortices, among others. These areas, which are distinctively involved in emotional and affect regulation of BPD patients, were the most informative regions to achieve both sensitivity and specificity values of 80% in SVM classification. The findings suggest that this new methodology can add clinical and potential diagnostic value to neuroimaging of psychiatric disorders. (C) 2012 Elsevier Ltd. All rights reserved.

Variance Reduction in Analytical Chemistry : New Numerical Methods in Chemometrics and Molecular Simulation

This thesis is based on five papers addressing variance reduction in different ways. The papers have in common that they all present new numerical methods. Paper I investigates quantitative structure-retention relationships from an image processing perspective, using an artificial neural network to preprocess three-dimensional structural descriptions of the studied steroid molecules. Paper II presents a new method for computing free energies. Free energy is the quantity that determines chemical equilibria and partition coefficients. The proposed method may be used for estimating, e.g., chromatographic retention without performing experiments. Two papers (III and IV) deal with correcting deviations from bilinearity by so-called peak alignment. Bilinearity is a theoretical assumption about the distribution of instrumental data that is often violated by measured data. Deviations from bilinearity lead to increased variance, both in the data and in inferences from the data, unless invariance to the deviations is built into the model, e.g., by the use of the method proposed in paper III and extended in paper IV. Paper V addresses a generic problem in classification; namely, how to measure the goodness of different data representations, so that the best classifier may be constructed. Variance reduction is one of the pillars on which analytical chemistry rests. This thesis considers two aspects on variance reduction: before and after experiments are performed. Before experimenting, theoretical predictions of experimental outcomes may be used to direct which experiments to perform, and how to perform them (papers I and II). After experiments are performed, the variance of inferences from the measured data are affected by the method of data analysis (papers III-V).

Prototipo de herramienta para el inventariado automático de piezas de Lego basado en visión artificial

Permitida la difusión del código bajo los términos de la licencia BSD de tres cláusulas.

The use of remote sensing imagery for monitoring recovery in the aftermath of a natural disaster: the Hurricane Katrina Case

Although Recovery is often defined as the less studied and documented phase of the Emergency Management Cycle, a wide literature is available for describing characteristics and sub-phases of this process. Previous works do not allow to gain an overall perspective because of a lack of systematic consistent monitoring of recovery utilizing advanced technologies such as remote sensing and GIS technologies. Taking into consideration the key role of Remote Sensing in Response and Damage Assessment, this thesis is aimed to verify the appropriateness of such advanced monitoring techniques to detect recovery advancements over time, with close attention to the main characteristics of the study event: Hurricane Katrina storm surge. Based on multi-source, multi-sensor and multi-temporal data, the post-Katrina recovery was analysed using both a qualitative and a quantitative approach. The first phase was dedicated to the investigation of the relation between urban types, damage and recovery state, referring to geographical and technological parameters. Damage and recovery scales were proposed to review critical observations on remarkable surge- induced effects on various typologies of structures, analyzed at a per-building level. This wide-ranging investigation allowed a new understanding of the distinctive features of the recovery process. A quantitative analysis was employed to develop methodological procedures suited to recognize and monitor distribution, timing and characteristics of recovery activities in the study area. Promising results, gained by applying supervised classification algorithms to detect localization and distribution of blue tarp, have proved that this methodology may help the analyst in the detection and monitoring of recovery activities in areas that have been affected by medium damage. The study found that Mahalanobis Distance was the classifier which provided the most accurate results, in localising blue roofs with 93.7% of blue roof classified correctly and a producer accuracy of 70%. It was seen to be the classifier least sensitive to spectral signature alteration. The application of the dissimilarity textural classification to satellite imagery has demonstrated the suitability of this technique for the detection of debris distribution and for the monitoring of demolition and reconstruction activities in the study area. Linking these geographically extensive techniques with expert per-building interpretation of advanced-technology ground surveys provides a multi-faceted view of the physical recovery process. Remote sensing and GIS technologies combined to advanced ground survey approach provides extremely valuable capability in Recovery activities monitoring and may constitute a technical basis to lead aid organization and local government in the Recovery management.

Pattern recognition analysis on heavy ion reaction data

One of the problems in the analysis of nucleus-nucleus collisions is to get information on the value of the impact parameter b. This work consists in the application of pattern recognition techniques aimed at associating values of b to groups of events. To this end, a support vec- tor machine (SVM) classifier is adopted to analyze multifragmentation reactions. This method allows to backtracing the values of b through a particular multidimensional analysis. The SVM classification con- sists of two main phase. In the first one, known as training phase, the classifier learns to discriminate the events that are generated by two different model:Classical Molecular Dynamics (CMD) and Heavy- Ion Phase-Space Exploration (HIPSE) for the reaction: 58Ni +48 Ca at 25 AMeV. To check the classification of events in the second one, known as test phase, what has been learned is tested on new events generated by the same models. These new results have been com- pared to the ones obtained through others techniques of backtracing the impact parameter. Our tests show that, following this approach, the central collisions and peripheral collisions, for the CMD events, are always better classified with respect to the classification by the others techniques of backtracing. We have finally performed the SVM classification on the experimental data measured by NUCL-EX col- laboration with CHIMERA apparatus for the previous reaction.

Evaluation of a multi variated analysis for the selection of t-tbar multijet events in the current CMS implemented environment at LHC

In the present study we are using multi variate analysis techniques to discriminate signal from background in the fully hadronic decay channel of ttbar events. We give a brief introduction to the role of the Top quark in the standard model and a general description of the CMS Experiment at LHC. We have used the CMS experiment computing and software infrastructure to generate and prepare the data samples used in this analysis. We tested the performance of three different classifiers applied to our data samples and used the selection obtained with the Multi Layer Perceptron classifier to give an estimation of the statistical and systematical uncertainty on the cross section measurement.

Exploring the use of local descriptors for fish recognition in LifeCLEF 2015

[EN]This paper summarizes the proposal made by the SIANI team for the LifeCLEF 2015 Fish task. The approach makes use of standard detection techniques, applying a multiclass SVM based classifier on large enough Regions Of Interest (ROIs) automatically extracted from the provided video frames. The selection of the detection and classification modules is based on the best performance achieved for the validation dataset consisting of 20 annotated videos. For that dataset, the best classification achieved for an ideal detection module, reaches an accuracy around 40%.

A study for the self similarity smile detection

[EN]This work makes an extensive experimental study of smile detection testing the Local Binary Patterns (LBP) combined with self similarity (LAC) as main descriptors of the image, along with the powerful Support Vector Machines classifier. Results show that error rates can be acceptable and the self similarity approach for the detection of smiles is suitable for real-time interaction, although there is still room for improvement.

Becoming visually familiar

Automatic face recognition has been mainly tackled by matching a new image to a set of previously computed identity models. The literature describes approximations where those identity models are based on a single sample or a set of them. However, face representation keeps being a topic of great debate in the psychology literature, with some results suggesting the use of an average image. In this paper, instead of restricting our system to a fixed and precomputed classifier, the system learns iteratively based on the experience extracted from each meeting.

Ottimizzazione delle reti di distribuzione idrica tramite algoritmi genetici multi-obiettivo

Water distribution networks optimization is a challenging problem due to the dimension and the complexity of these systems. Since the last half of the twentieth century this field has been investigated by many authors. Recently, to overcome discrete nature of variables and non linearity of equations, the research has been focused on the development of heuristic algorithms. This algorithms do not require continuity and linearity of the problem functions because they are linked to an external hydraulic simulator that solve equations of mass continuity and of energy conservation of the network. In this work, a NSGA-II (Non-dominating Sorting Genetic Algorithm) has been used. This is a heuristic multi-objective genetic algorithm based on the analogy of evolution in nature. Starting from an initial random set of solutions, called population, it evolves them towards a front of solutions that minimize, separately and contemporaneously, all the objectives. This can be very useful in practical problems where multiple and discordant goals are common. Usually, one of the main drawback of these algorithms is related to time consuming: being a stochastic research, a lot of solutions must be analized before good ones are found. Results of this thesis about the classical optimal design problem shows that is possible to improve results modifying the mathematical definition of objective functions and the survival criterion, inserting good solutions created by a Cellular Automata and using rules created by classifier algorithm (C4.5). This part has been tested using the version of NSGA-II supplied by Centre for Water Systems (University of Exeter, UK) in MATLAB® environment. Even if orientating the research can constrain the algorithm with the risk of not finding the optimal set of solutions, it can greatly improve the results. Subsequently, thanks to CINECA help, a version of NSGA-II has been implemented in C language and parallelized: results about the global parallelization show the speed up, while results about the island parallelization show that communication among islands can improve the optimization. Finally, some tests about the optimization of pump scheduling have been carried out. In this case, good results are found for a small network, while the solutions of a big problem are affected by the lack of constraints on the number of pump switches. Possible future research is about the insertion of further constraints and the evolution guide. In the end, the optimization of water distribution systems is still far from a definitive solution, but the improvement in this field can be very useful in reducing the solutions cost of practical problems, where the high number of variables makes their management very difficult from human point of view.

Kernel Methods for Tree Structured Data

Machine learning comprises a series of techniques for automatic extraction of meaningful information from large collections of noisy data. In many real world applications, data is naturally represented in structured form. Since traditional methods in machine learning deal with vectorial information, they require an a priori form of preprocessing. Among all the learning techniques for dealing with structured data, kernel methods are recognized to have a strong theoretical background and to be effective approaches. They do not require an explicit vectorial representation of the data in terms of features, but rely on a measure of similarity between any pair of objects of a domain, the kernel function. Designing fast and good kernel functions is a challenging problem. In the case of tree structured data two issues become relevant: kernel for trees should not be sparse and should be fast to compute. The sparsity problem arises when, given a dataset and a kernel function, most structures of the dataset are completely dissimilar to one another. In those cases the classifier has too few information for making correct predictions on unseen data. In fact, it tends to produce a discriminating function behaving as the nearest neighbour rule. Sparsity is likely to arise for some standard tree kernel functions, such as the subtree and subset tree kernel, when they are applied to datasets with node labels belonging to a large domain. A second drawback of using tree kernels is the time complexity required both in learning and classification phases. Such a complexity can sometimes prevents the kernel application in scenarios involving large amount of data. This thesis proposes three contributions for resolving the above issues of kernel for trees. A first contribution aims at creating kernel functions which adapt to the statistical properties of the dataset, thus reducing its sparsity with respect to traditional tree kernel functions. Specifically, we propose to encode the input trees by an algorithm able to project the data onto a lower dimensional space with the property that similar structures are mapped similarly. By building kernel functions on the lower dimensional representation, we are able to perform inexact matchings between different inputs in the original space. A second contribution is the proposal of a novel kernel function based on the convolution kernel framework. Convolution kernel measures the similarity of two objects in terms of the similarities of their subparts. Most convolution kernels are based on counting the number of shared substructures, partially discarding information about their position in the original structure. The kernel function we propose is, instead, especially focused on this aspect. A third contribution is devoted at reducing the computational burden related to the calculation of a kernel function between a tree and a forest of trees, which is a typical operation in the classification phase and, for some algorithms, also in the learning phase. We propose a general methodology applicable to convolution kernels. Moreover, we show an instantiation of our technique when kernels such as the subtree and subset tree kernels are employed. In those cases, Direct Acyclic Graphs can be used to compactly represent shared substructures in different trees, thus reducing the computational burden and storage requirements.

Progetto di reti Sensori Wireless e tecniche di Fusione Sensoriale

Ambient Intelligence (AmI) envisions a world where smart, electronic environments are aware and responsive to their context. People moving into these settings engage many computational devices and systems simultaneously even if they are not aware of their presence. AmI stems from the convergence of three key technologies: ubiquitous computing, ubiquitous communication and natural interfaces. The dependence on a large amount of fixed and mobile sensors embedded into the environment makes of Wireless Sensor Networks one of the most relevant enabling technologies for AmI. WSN are complex systems made up of a number of sensor nodes, simple devices that typically embed a low power computational unit (microcontrollers, FPGAs etc.), a wireless communication unit, one or more sensors and a some form of energy supply (either batteries or energy scavenger modules). Low-cost, low-computational power, low energy consumption and small size are characteristics that must be taken into consideration when designing and dealing with WSNs. In order to handle the large amount of data generated by a WSN several multi sensor data fusion techniques have been developed. The aim of multisensor data fusion is to combine data to achieve better accuracy and inferences than could be achieved by the use of a single sensor alone. In this dissertation we present our results in building several AmI applications suitable for a WSN implementation. The work can be divided into two main areas: Multimodal Surveillance and Activity Recognition. Novel techniques to handle data from a network of low-cost, low-power Pyroelectric InfraRed (PIR) sensors are presented. Such techniques allow the detection of the number of people moving in the environment, their direction of movement and their position. We discuss how a mesh of PIR sensors can be integrated with a video surveillance system to increase its performance in people tracking. Furthermore we embed a PIR sensor within the design of a Wireless Video Sensor Node (WVSN) to extend its lifetime. Activity recognition is a fundamental block in natural interfaces. A challenging objective is to design an activity recognition system that is able to exploit a redundant but unreliable WSN. We present our activity in building a novel activity recognition architecture for such a dynamic system. The architecture has a hierarchical structure where simple nodes performs gesture classification and a high level meta classifiers fuses a changing number of classifier outputs. We demonstrate the benefit of such architecture in terms of increased recognition performance, and fault and noise robustness. Furthermore we show how we can extend network lifetime by performing a performance-power trade-off. Smart objects can enhance user experience within smart environments. We present our work in extending the capabilities of the Smart Micrel Cube (SMCube), a smart object used as tangible interface within a tangible computing framework, through the development of a gesture recognition algorithm suitable for this limited computational power device. Finally the development of activity recognition techniques can greatly benefit from the availability of shared dataset. We report our experience in building a dataset for activity recognition. Such dataset is freely available to the scientific community for research purposes and can be used as a testbench for developing, testing and comparing different activity recognition techniques.

Body ares networks for human motor assessment

Healthcare, Human Computer Interfaces (HCI), Security and Biometry are the most promising application scenario directly involved in the Body Area Networks (BANs) evolution. Both wearable devices and sensors directly integrated in garments envision a word in which each of us is supervised by an invisible assistant monitoring our health and daily-life activities. New opportunities are enabled because improvements in sensors miniaturization and transmission efficiency of the wireless protocols, that achieved the integration of high computational power aboard independent, energy-autonomous, small form factor devices. Application’s purposes are various: (I) data collection to achieve off-line knowledge discovery; (II) user notification of his/her activities or in case a danger occurs; (III) biofeedback rehabilitation; (IV) remote alarm activation in case the subject need assistance; (V) introduction of a more natural interaction with the surrounding computerized environment; (VI) users identification by physiological or behavioral characteristics. Telemedicine and mHealth [1] are two of the leading concepts directly related to healthcare. The capability to borne unobtrusiveness objects supports users’ autonomy. A new sense of freedom is shown to the user, not only supported by a psychological help but a real safety improvement. Furthermore, medical community aims the introduction of new devices to innovate patient treatments. In particular, the extension of the ambulatory analysis in the real life scenario by proving continuous acquisition. The wide diffusion of emerging wellness portable equipment extended the usability of wearable devices also for fitness and training by monitoring user performance on the working task. The learning of the right execution techniques related to work, sport, music can be supported by an electronic trainer furnishing the adequate aid. HCIs made real the concept of Ubiquitous, Pervasive Computing and Calm Technology introduced in the 1988 by Marc Weiser and John Seeley Brown. They promotes the creation of pervasive environments, enhancing the human experience. Context aware, adaptive and proactive environments serve and help people by becoming sensitive and reactive to their presence, since electronics is ubiquitous and deployed everywhere. In this thesis we pay attention to the integration of all the aspects involved in a BAN development. Starting from the choice of sensors we design the node, configure the radio network, implement real-time data analysis and provide a feedback to the user. We present algorithms to be implemented in wearable assistant for posture and gait analysis and to provide assistance on different walking conditions, preventing falls. Our aim, expressed by the idea to contribute at the development of a non proprietary solutions, driven us to integrate commercial and standard solutions in our devices. We use sensors available on the market and avoided to design specialized sensors in ASIC technologies. We employ standard radio protocol and open source projects when it was achieved. The specific contributions of the PhD research activities are presented and discussed in the following. • We have designed and build several wireless sensor node providing both sensing and actuator capability making the focus on the flexibility, small form factor and low power consumption. The key idea was to develop a simple and general purpose architecture for rapid analysis, prototyping and deployment of BAN solutions. Two different sensing units are integrated: kinematic (3D accelerometer and 3D gyroscopes) and kinetic (foot-floor contact pressure forces). Two kind of feedbacks were implemented: audio and vibrotactile. • Since the system built is a suitable platform for testing and measuring the features and the constraints of a sensor network (radio communication, network protocols, power consumption and autonomy), we made a comparison between Bluetooth and ZigBee performance in terms of throughput and energy efficiency. Test in the field evaluate the usability in the fall detection scenario. • To prove the flexibility of the architecture designed, we have implemented a wearable system for human posture rehabilitation. The application was developed in conjunction with biomedical engineers who provided the audio-algorithms to furnish a biofeedback to the user about his/her stability. • We explored off-line gait analysis of collected data, developing an algorithm to detect foot inclination in the sagittal plane, during walk. • In collaboration with the Wearable Lab – ETH, Zurich, we developed an algorithm to monitor the user during several walking condition where the user carry a load. The remainder of the thesis is organized as follows. Chapter I gives an overview about Body Area Networks (BANs), illustrating the relevant features of this technology and the key challenges still open. It concludes with a short list of the real solutions and prototypes proposed by academic research and manufacturers. The domain of the posture and gait analysis, the methodologies, and the technologies used to provide real-time feedback on detected events, are illustrated in Chapter II. The Chapter III and IV, respectively, shown BANs developed with the purpose to detect fall and monitor the gait taking advantage by two inertial measurement unit and baropodometric insoles. Chapter V reports an audio-biofeedback system to improve balance on the information provided by the use centre of mass. A walking assistant based on the KNN classifier to detect walking alteration on load carriage, is described in Chapter VI.

Informationsgehalt von ERS-1/-2 SAR-Daten zur Erfassung der Arteninventare und des Zustandes landwirtschaftlich genutzter Böden und Vegetation

Auf einer drei Anbauperioden umfassenden Ground Truth Datenbasis wird der Informationsgehalt multitemporaler ERS-1/-2 Synthetic Aperture Radar (SAR) Daten zur Erfassung der Arteninventare und des Zustandes landwirtschaftlich genutzter Böden und Vegetation in Agrarregionen Bayerns evaluiert.Dazu wird ein für Radardaten angepaßtes, multitemporales, auf landwirtschaftlichen Schlägen beruhendes Klassifizierungsverfahren ausgearbeitet, das auf bildstatistischen Parametern der ERS-Zeitreihen beruht. Als überwachte Klassifizierungsverfahren wird vergleichend der Maximum-Likelihood-Klassifikator und ein Neuronales-Backpropagation-Netz eingesetzt. Die auf Radarbildkanälen beruhenden Gesamtgenauigkeiten variieren zwischen 75 und 85%. Darüber hinaus wird gezeigt, daß die interferometrische Kohärenz und die Kombination mit Bildkanälen optischer Sensoren (Landsat-TM, SPOT-PAN und IRS-1C-PAN) zur Verbesserung der Klassifizierung beitragen. Gleichermaßen können die Klassifizierungsergebnisse durch eine vorgeschaltete Grobsegmentierung des Untersuchungsgebietes in naturräumlich homogene Raumeinheiten verbessert werden. Über die Landnutzungsklassifizierung hinaus, werden weitere bio- und bodenphysikalische Parameter aus den SAR-Daten anhand von Regressionsmodellen abgeleitet. Im Mittelpunkt stehen die Paramter oberflächennahen Bodenfeuchte vegetationsfreier/-armer Flächen sowie die Biomasse landwirtschaftlicher Kulturen. Die Ergebnisse zeigen, daß mit ERS-1/-2 SAR-Daten eine Messung der Bodenfeuchte möglich ist, wenn Informationen zur Bodenrauhigkeit vorliegen. Hinsichtlich der biophysikalischen Parameter sind signifikante Zusammenhänge zwischen der Frisch- bzw. Trockenmasse des Vegetationsbestandes verschiedener Getreide und dem Radarsignal nachweisbar. Die Biomasse-Informationen können zur Korrektur von Wachstumsmodellen genutzt werden und dazu beitragen, die Genauigkeit von Ertragsschätzungen zu steigern.

Numeralklassifikatoren im Thai

Wie viele andere Sprachen Ost- und Südostasiens ist das Thai eine numerusneutrale Sprache, in der ein Nomen lediglich das Konzept benennt und keinen Hinweis auf die Anzahl der Objekte liefert. Um Nomina im Thai zählen zu können, ist der Klassifikator (Klf) nötig, der die Objekte anhand ihrer semantischen Schlüsseleigenschaft herausgreift und individualisiert. Neben der Klassifikation stellt die Individualisierung die Hauptfunktion des Klf dar. Weitere Kernfunktionen des Klf außerhalb des Zählkontextes sind die Markierung der Definitheit, des Numerus sowie des Kontrasts. Die wichtigsten neuen Ergebnisse dieser Arbeit, die sowohl die Ebenen der Grammatik und Semantik als auch die der Logik und Pragmatik integriert, sind folgende: Im Thai kann der Klf sowohl auf der Element- als auch auf der Mengenebene agieren. In der Verbindung mit einem Demonstrativ kann der Klf auch eine pluralische Interpretation hervorrufen, wenn er auf eine als pluralisch präsupponierte Gesamtmenge referiert oder die Gesamtmenge in einer Teil-Ganzes-Relation individualisiert. In einem Ausdruck, der bereits eine explizite Zahlangabe enthält, bewirkt die Klf-Demonstrativ-Konstruktion eine Kontrastierung von Mengen mit gleichen Eigenschaften. Wie auch der Individualbegriff besitzt der Klf Intension und Extension. Intension und Extension von Thai-Klf verhalten sich umgekehrt proportional, d.h. je spezifischer der Inhalt eines Klf ist, desto kleiner ist sein Umfang. Der Klf signalisiert das Schlüsselmerkmal, das mit der Intension des Nomens der Identifizierung des Objekts dient. Der Klf individualisiert das Nomen, indem er Teilmengen quantifiziert. Er kann sich auf ein Objekt, eine bestimmte Anzahl von Objekten oder auf alle Objekte beziehen. Formal logisch lassen sich diese Funktionen mithilfe des Existenz- und des Allquantors darstellen. Auch die Nullstelle (NST) läßt sich formal logisch darstellen. Auf ihren jeweiligen Informationsgehalt reduziert, ergeben sich für Klf und NST abhängig von ihrer Positionierung verschiedene Informationswerte: Die Opposition von Klf und NST bewirkt in den Fragebögen ausschließlich skalare Q-Implikaturen, die sich durch die Informationsformeln in Form einer Horn-Skala darstellen lassen. In einem sich aufbauenden Kontext transportieren sowohl Klf als auch NST in der Kontextmitte bekannte Informationen, wodurch Implikaturen des M- bzw. I-Prinzips ausgelöst werden. Durch die Verbindung der Informationswerte mit den Implikaturen des Q-, M- und I-Prinzips lässt sich anhand der Positionierung direkt erkennen, wann der Klf die Funktion der Numerus-, der Definitheits- oder der Kontrast-Markierung erfüllt.