This thesis describes the development and validation of a predictive model of graphical user interface (GUI) aesthetics. The development was informed by the processing-fluency theory of aesthetic pleasure and involved outlining several visual dimensions of GUI designs, which could affect aesthetics impression. Each of the dimensions was grounded in theory and represents a unique visual aspect of GUI design. The resulting model automatically evaluates the design dimensions and combines them in an estimate of the average impression that GUI appearance would make on the user population. The model was validated in a number of user studies proving high validity and reliability. The model outputs an aesthetics score ofuser impression and could inform the creation of more beautiful GUIs by highlighting which of the design dimensions could be improved. The thesis describes the studies that validated the model on several types of GUIs and demonstrated a potential application of the model in future research and practice.

The retrieval of geo-/bio-physical variables from remote sensing imagery is a challenging and important research field. On the one hand, advances in electronics, engineering and space sciences are offering to the users community new sensors capable to acquire information on the Earth surface with higher accuracy and improved features with respect to the past. On the other hand, the need of large-scale, accurate and up-to-date mapping and monitoring of natural targets and physical processes is becoming fundamental for many application domains. This calls for the development of accurate, robust and effective retrieval methodologies. The main goal of this thesis is to investigate and develop advanced methods and systems for the retrieval of geo-/bio-physical variables from satellite remote sensing imagery being able to exploit the potential of new and upcoming satellite systems and support real application domains. Special attention has been devoted to the definition of methods and to the analysis of data acquired in the challenging mountain environment. The activity carried out and presented in this dissertation is oriented to investigate the main limitations of the existing methodologies for addressing the estimation problem and to develop novel and improved systems that can overcome the drawbacks identified. In particular, the following main novel contributions are proposed in this thesis: a) A theoretical and empirical comparative analysis of non-linear machine learning regression methods, namely the Multi-Layer Perceptron Neural Network and the Support Vector Regression, for soil moisture retrieval in different operational scenarios. b) A novel multi-objective model-selection strategy for tuning the free parameters of non-linear regression methods taking into account different quality metrics that are jointly optimized. c) A novel hybrid approach to the retrieval of geo-/bio-physical variables from remote sensing data integrating both theoretical electromagnetic models and field reference measurements. d) A sensitivity analysis and a retrieval system for soil moisture content estimation from new generation SAR imagery in an Alpine catchment. e) An empirical study on the effectiveness of fully-polarimetric SAR signals for soil moisture estimation in mountain areas. f) An improved algorithm for mapping and monitoring Green Area Index (GAI) in Alpine pastures and meadows from satellite MODIS imagery. Qualitative and quantitative experimental results obtained on real remotely sensed data confirm the effectiveness of the proposed solutions.

Languages are well known to be diverse on all structural levels, from the smallest (phonemic) to the broadest (pragmatic). We propose a set of formal, quantitative measures for the language diversity of linguistic phenomena, the resource incompleteness, and resource incorrectness. We apply all these measures to lexical semantics where we show how evidence of a high degree of universality within a given language set can be used to extend lexico-semantic resources in a precise, diversity-aware manner. We demonstrate our approach on several case studies: First is on polysemes and homographs among cases of lexical ambiguity. Contrarily to past research that focused solely on exploiting systematic polysemy, the notion of universality provides us with an automated method also capable of predicting irregular polysemes. Second is to automatically identify cognates from the existing lexical resource across different orthographies of genetically unrelated languages. Contrarily to past research that focused on detecting cognates from 225 concepts of Swadesh list, we captured 3.1 million cognate pairs across 40 different orthographies and 335 languages by exploiting the existing wordnet-like lexical resources.

Deep learning is an emerging technology that is considered one of the most promising directions for reaching higher levels of artificial intelligence. Among the other achievements, building computers that understand speech represents a crucial leap towards intelligent machines. Despite the great efforts of the past decades, however, a natural and robust human-machine speech interaction still appears to be out of reach, especially when users interact with a distant microphone in noisy and reverberant environments. The latter disturbances severely hamper the intelligibility of a speech signal, making Distant Speech Recognition (DSR) one of the major open challenges in the field. This thesis addresses the latter scenario and proposes some novel techniques, architectures, and algorithms to improve the robustness of distant-talking acoustic models. We first elaborate on methodologies for realistic data contamination, with a particular emphasis on DNN training with simulated data. We then investigate on approaches for better exploiting speech contexts, proposing some original methodologies for both feed-forward and recurrent neural networks. Lastly, inspired by the idea that cooperation across different DNNs could be the key for counteracting the harmful effects of noise and reverberation, we propose a novel deep learning paradigm called “network of deep neural networks”. The analysis of the original concepts were based on extensive experimental validations conducted on both real and simulated data, considering different corpora, microphone configurations, environments, noisy conditions, and ASR tasks.

O objetivo da dissertação é examinar o dever de divulgar fato relevante imposto aos administradores de companhia aberta pela Lei nº 6.404/76. Para tanto, na primeira parte do trabalho analisa-se, sob uma perspectiva ampla, o dever de informar na sociedade anônima e no mercado de capitais, abrangendo os fundamentos para a imposição de tais deveres, sua relação com as regras disciplinadoras da informação nas sociedades anônimas em geral e com as aplicáveis às companhias abertas em particular, bem como a estrutura da disciplina legal, regulamentar e autorregulamentar. A segunda parte do trabalho destina-se a investigar o conceito de fato relevante, definido pela lei como aquele que possa influir, de modo ponderável, na decisão dos investidores do mercado de vender ou comprar valores mobiliários emitidos pela companhia. As origens do conceito são identificadas no direito norte-americano, que inspirou também o seu tratamento na legislação comunitária europeia e em outros países. O conceito na legislação brasileira é analisado a partir de seus elementos constitutivos: o investidor de mercado (e quais pessoas devem ser consideradas como tal), a influência ponderável sobre a decisão de investimento (especialmente se a influência é apenas potencial ou deve ser efetiva) e a noção de ato ou fato ocorridos nos negócios da companhia (e de que modo o conceito exclui atos que, não obstante possam influenciar as decisões de investimento, não ocorrem nos negócios sociais). A terceira parte do trabalho analisa o dever de divulgar fato relevante, considerando a responsabilidade pela divulgação, a forma de divulgação e as exceções à regra geral de divulgação imediata, bem como as consequências de seu descumprimento nas esferas administrativa, civil e criminal. / Lobiettivo della dissertazione è esaminare il dovere di divulgare un fatto rilevante imposto agli amministratori di società per azioni quotate in Borsa dalla Legge nº 6.404/76. Perciò, vengono analizzati, nella prima parte del lavoro, sotto unampia prospettiva, il dovere di informare nella società anonima e nel mercato di capitali, comprensivo di fondamenti per limposizione di tali doveri, la loro relazione con le regole relative allinformazione nelle società anonime in genere e con quelle applicabili alle società per azioni quotate in Borsa in particolare, nonché la struttura della disciplina legale, regolamentare e autoregolamentare. La seconda parte del lavoro è destinata a investigare il concetto di fatto rilevante, definito dalla legge come quello che può influire, in modo ponderabile, sulla decisione degli investitori del mercato di vendere o comprare valori mobiliari emessi dalla società. Le origini del concetto sono identificati nel diritto americano, che ha anche ispirato il suo trattamento nella legislazione comunitaria europea e in altri paesi. Il concetto nella legislazione brasiliana è analizzato in base ai suoi elementi costitutivi: linvestitore di mercato (e quali persone devono essere considerate come tali a questo fine), linfluenza ponderabile sulla decisione dinvestimento (specialmente se linfluenza è soltanto potenziale o deve essere effettiva) e la nozione di atto o fatto avvenuti negli affari della società (e in che modo questo concetto esclude atti che, nonostante possano influenzare le decisioni di investimento, non sono avvenuti negli affari sociali). La terza parte del lavoro analizza il dovere di divulgare un fatto rilevante, considerando la responsabilità della divulgazione, la forma di divulgazione e le eccezioni alla regola generale di divulgazione immediata, così come le conseguenze di suo inadempimento, nella sfera amministrativa, civile e criminale.

We live in exciting times. The fast paced growth in mobile computers has put powerful computational devices in the palm of our hands. Blazing fast connectivity has made human-human, human-machine, and machine-machine communication effortless. Wearable devices and the internet of things have made monitoring every aspect of our lives easier. This has given rise to the domain of quantified self where we can continuous record and quantify the various signals generated in everyday life. Sensors on smartphones can continuously record our location and motion profile. Sensors on wearable devices can track changes in our bodies’ physiological responses. This monitoring also has the capability to revolutionise the health care domain by creating more informed and involved patients. This has the potential to shift care-management from a physician-centric approach to a patient-centric approach allowing individuals to create more empowered patients and individuals who are in better control of their health. However, the data deluge from all these sources can sometimes be overwhelming. There is a need for intelligent technology that can help us navigate the data and take informed decisions. The goal of this work is to develop a mobile, personal intelligent agent platform that can become a digital companion to live with the user. It can monitor the covert and overt signal streams of the user, identify activity and stress levels to help the users’ make healthy choices regarding their lives. This thesis particularly targets patients suffering from or at-risk of essential hypertension since its a difficult condition to detect and manage. This thesis delivers the following contributions: 1) An intelligent personal agent platform for on-the-go continuous monitoring of covert and overt signals. 2) A machine learning algorithm for accurate recognition of activities using smartphone signals recorded from in-the-wild scenarios. 3) A machine learning pipeline to combine various physiological signal streams, motion profiles, and user annotations for on-the-go stress recognition. 4) We design and train a complete signal processing and classification system for hypertension prediction. 5) Through a small pilot study we demonstrate that this system can distinguish between hypertensive and normotensive subjects with high accuracy.

Idea Management (IM) is the process of requesting, collecting, selecting and evaluating ideas to develop new and innovative products, services or regulations, or to improve existing ones. The process is supported by dedicated Idea Management systems (IMS), which lets people propose ideas, as well as rate and place comments on other users’ suggestions. When used in the civic domain, IM serves as a tool to engage citizens in processes of innovation of public services, laws, and regulations. A key ingredient in the success of IM is the community of participants. The larger the community, the more diverse views are likely to appear and diversity of views increases the chances of discovering valuable ideas that can lead to innovations. However, having a large number of people participating in IMS is a hard challenge; it requires an understanding of the people and their needs and designing the technology to match these characteristics. In this thesis, we aim at involving the society at large into IM processes. Achieving this ambitious goal requires integrating IMS with people’s everyday life tools and spaces of participation. We understand that tools for civic engagement should engage people on their own terms and should be readily available. We meet these requirements by proposing an approach that integrates IMS into common physical and virtual spaces of participation enabling people to participate in IM using ordinary tools and without having to step outside their daily habits. In a systematic and extensive study of the literature about technologies used to foster civic engagement in innovation processes, we found that the choice of technology and its “situated- ness” is essential in granting ease of public access and promoting inclusive processes of civic engagement. We also discovered that civic engagement technologies still have room to improve their use of multiple channels of participation. In this regard, we saw social networking sites such as Facebook and Twitter as having a strong potential to lower participation barriers and engage citizens, considering how pervasive these sites are today as daily tools. We show how the lessons learned can be applied in practice by presenting two solutions to increase participation in IMS. The first solution is a platform that extends IMS by integrating them into displays located in public spaces. From this experience, we found that taking the right instruments to where people actually are is important to address specific inequalities regarding access to technology. We also saw that the display represented for citizens not only an opportunity to make their voice being heard but also an occasion for socialization. The second solution is a model and tool that empower IMS through Facebook services. Here we found that the integration with Facebook facilitated participation by reducing the friction related to getting informed and involved in IM. Also, the participants reported that the familiarity and easy to use of Facebook features represented an advantage for participation. We informed the design of both solutions with large- and medium-scale data analysis studies on the behavior (individual and collective), practices, and motivation factors of IM communities’ participants.

Nowadays, the field of wheeled robotics is undergoing an impressive growth and development. Different hardware and software components are being developed and applied in various contexts, including assistive robotics, industrial robotics, automotive, ... Motion Planning is a fundamental aspect for the development of autonomous wheeled mobile robots. The capability of planning safe, smooth trajectories, and to locally adjust them in real-time to deal with contingent situations and avoid collisions is an essential requirement to allow robots to work and perform activities in public spaces shared with humans. Moreover, in general, efficiency is a key constraint for this kind of applications, given the limited computational power usually available on robotic platforms. In this thesis, we focus on the development of efficient algorithms to solve different kind of motion planning problems. Specifically, in the first part of the thesis, we propose a complete planning system for an assisitive robot supporting the navigation of older users. The developed planner generates paths connecting different locations on the map, that are smooth and specifically tailored to optimize the comfort perceived by the human users. During the navigation, the system applies an efficient model to predict the behaviours of the surrounding pedestrians, and to locally adapt the reference path to minimise the probability of collisions. Finally, the motion planner is integrated with an "high-level" reasoning component, to generate and propose complete activities, like the visit to a museum or a shopping mall, specifically tailored to the preferences, needs and requirements of each user. In the second part of the thesis, we show how the efficient solutions and building blocks developed for the assistive robots, can be adapted and applied also to a completely different context, such as the generation of optimal trajectories for an autonomous racing vehicle.

Fast development of information and communication technologies made available vast amounts of heterogeneous information. With these amounts growing faster and faster, information integration and search technologies are becoming a key for the success of information society. To handle such amounts efficiently, data needs to be leveraged and analysed at deep levels. Metadata is a traditional way of getting leverage over the data. Deeper levels of analysis include language analysis, starting from purely string-based (keyword) approaches, continuing with syntactic-based approaches and now semantics is about to be included in the processing loop. Metadata gives a leverage over the data. Often a natural language, being the easiest way of expression, is used in metadata. We call such metadata ``natural language metadata''. The examples include various titles, captions and labels, such as web directory labels, picture titles, classification labels, business directory category names. These short pieces of text usually describe (sets of) objects. We call them ``descriptive phrases''. This thesis deals with a problem of understanding natural language metadata for its further use in semantics aware applications. This thesis contributes by portraying descriptive phrases, using the results of analysis of several collected and annotated datasets of natural language metadata. It provides an architecture for the natural language metadata understanding, complete with the algorithms and the implementation. This thesis contains the evaluation of the proposed architecture.

State-of-the-art neural machine translation systems typically have low accuracy in translating rare or unseen words due to the requirement of using a fixed-size word vocabulary during training. In addition to controlling the model complexity, this limitation is also related to the difficulty of learning accurate word representations under conditions of high data sparsity. This problem is an important bottleneck on performance, especially in morphologically-rich languages, where the word vocabulary tends to be huge and sparse. In this dissertation, we propose to solve the vocabulary limitation problem in neural machine translation by integrating morphology learning within the translation model, aiding to learn richer word representations in terms of phonological and morphological information. Our model improves the accuracy while translating into low-resource and morphologically-rich languages and shows better generalization capability over varieties of languages with different morphological characteristics.