In the field of Ubiquitous Computing, a significant problem of building accurate machine learning models is the effort and time consuming process to gather labeled data for the learning algorithm. Moreover, efficient data use demands are constantly growing. These demands for efficient data use are growing constantly. Researchers are therefore exploring the use of machine learning techniques to overcome the problem of data scarcity. In healthcare, classification tasks require a ground truth normally provided by an expert physician, ending up with a small set of labeled data with a larger set of unlabeled data. It is also common to rely on self-reported data through questionnaires, however, this introduce an extra burden to the user who is not always able or willing to fill in. Finally, in some healthcare domains it is important to be able to provide immediate response (feedback), even if the user is not familiarized with the use of an application. In all of these cases the amount of available data may be insufficient to produce reliable models. This thesis proposes a new approach specifically designed for the challenges in producing better predictive models. We propose using our novel Intermediate Models to predict the mood variables associated with the questionnaire using data acquired from smartphones. Then, we use the predicted mood variables with the rest of the data to predict the class, in our empirical assessment, the state mood of a bipolar disorder patient or stress levels of employees have been used. The motivation behind this new approach is that there are relevant proposed methods such as latent variables used as intermediate information helping to create better predictive models. These methods are used in literature to complete the missing data using the most common value, the most probable value given the class, or induce a model for predicting missing values using all the information from features and the class. However, these variables are artificially created and used as intermediate information to build better model. In our Intermediate Models, we know in advance how many mood variables to use and we have the information from these variables, which allow us to produce better models. To address scarce data, we propose applying a semi-supervised learning setting while taking advantage of the presence of all unlabeled datasets. In addition, we propose using transfer learning methods that is used to improve the learning performance with the aim at avoiding expensive data labeling efforts. To the best of our knowledge, there are few works that have used transfer learning for healthcare applications to address the problem of limited labeled data. The proposed methods have been applied in two different healthcare fields: mental-health and human behaviour field. This thesis addresses two classification problems, a) classification of episodic state of bipolar disorder patients, and b) detecting work-related stress using data acquired from smartphone sensing modalities.

Preference-based decision problems often involve choosing one among a large set of options, making common tasks like buying a car or a domestic appliance very challenging for a customer to handle on her own. This is especially true when buying online, where the amount of available options is humongous, and expert advice is yet limited. Recommender systems have become essential computational tools for aiding users in this endeavor. Recommender systems represent one of the most successful applications of artificial intelligence. In the last decades, several recommendation approaches have been proposed for different types of applications, from assisted browsing of product catalogs to personalization of results in search engines. Depending on the application, the job of the recommender system may be to recommend a satisfying option for the given context, as in finding the next best song to play, as opposed to helping the user in finding an optimal instance, e.g. when looking for an apartment. The former is generally handled by data-driven approaches, such as collaborative filtering and contextual bandits, while in the latter case data is usually scarce, making it necessary to employ specialized algorithms for preference elicitation. Preference elicitation algorithms interactively build a utility model of the user preferences and then recommend the instances with the highest utility. Preference elicitation is especially effective when recommending infrequently purchased items, such as professional work tools, electronic devices and other products that can be explicitly stored e.g. in the database of an e-commerce website. Standard preference elicitation algorithms, however, struggle when the options are so numerous that cannot even be explicitly enumerated, and instead need to be represented implicitly as a collection of variables and constraints. Indeed, when a customer wants to configure a product by putting several components together, e.g. for a custom personal computer, the option space is combinatorial and grows exponentially with the number of components, making it impractical to store every single feasible combination explicitly. This is an example of constructive decision problem, in which an object has to be synthesized on the basis of the preferences of the customer and the constraints over the configuration domain. Constructive problems such as product configuration have traditionally been addressed by specialized configurator systems, which guide the user through the configuration process component by component and check whether the user choices are consistent with the set of feasibility constraints. Over the years, however, the limitations of product configurators for mass customization have become apparent. With the growing scale of configuration problems, product configurators have become more difficult for non-experts to use and ultimately do not provide relief against the "mass confusion" caused by the sheer amount of choice. Research in this field has progressively been integrating recommendation technologies into configuration systems, in order to make them more flexible and easy to use. Preference elicitation in product configuration has been attempted as well but still remains a challenge. We propose a generic framework for preference elicitation in constructive domains, that is able to scale to large combinatorial problems better than existing techniques. Our constructive preference elicitation framework is based on online structured prediction, a machine learning technique that deals with sequential decision problems over structured objects. By combining online structured prediction and state-of-the-art constraint solvers we can efficiently learn user utility models and make increasingly better recommendations for complex preference-based constructive problems such as product configuration. In particular, we favor the use of coactive learning, an online structured prediction framework for preference learning. Coactive learning is particularly well suited for constructive preference elicitation as it may be seen as a cooperation between the user and the system. The user and the systems interact through "coactive" feedback: after each recommendation, the user provides a modification that makes it slightly better for her preferences. This type of feedback is very flexible and can be acquired both explicitly and implicitly from the user actions. Coactive learning also comes with theoretical convergence guarantees and a set of ready-made extensions for many related problems such as learning in a multi-user setting and learning with approximate constraint solvers. In this thesis we detail our coactive learning approach to constructive preference elicitation, and propose extensions for scaling up to very large constructive problems and personalizing the utility model. We also applied our framework to two important classes of constructive preference elicitation problems, namely layout synthesis and product bundling. The former is a design process for arranging objects into a given space, while the latter is a kind of product configuration problem in which the object to configure is a package of different products and services. Within the product bundling application, we also performed an extensive validation involving real participants, which highlights the practical benefits of our approach.

The various time-frequency (TF) representations of acoustic signals share the common objective to describe the temporal evolution of the spectral content of the signal, i.e., how the energy, or intensity, of the signal is changing in time. Many TF representations have been proposed in the past, and among them the short-time Fourier transform (STFT) is the one most commonly found in the core of acoustic signal processing techniques. However, certain problems that arise from the use of the STFT have been extensively discussed in the literature. These problems concern the unavoidable trade-off between the time and frequency resolution, and the fact that the selected resolution is fixed over the whole spectrum. In order to improve upon the spectrogram, several variations have been proposed over the time. One of these variations, stems from a promising method called reassignment. According to this method, the traditional spectrogram, as obtained from the STFT, is reassigned to a sharper representation called the Reassigned Spectrogram (RS). In this thesis we elaborate on approaches that utilize the RS as the TF representation of acoustic signals, and we exploit this representation in the context of different applications, as for instance speech recognition and melody extraction. The first contribution of this work is a method for speech parametrization, which results in a set of acoustic features called time-frequency reassigned cepstral coefficients (TFRCC). Experimental results show the ability of TFRCC features to present higher level characteristics of speech, a fact that leads to advantages in phone-level speech segmentation and speech recognition. The second contribution is the use of the RS as the basis to extract objective quality measures, and in particular the reassigned cepstral distance and the reassigned point-wise distance. Both measures are used for channel selection (CS), following our proposal to perform objective quality measure based CS for improving the accuracy of speech recognition in a multi-microphone reverberant environment. The final contribution of this work, is a method to detect harmonic pitch contours from singing voice signals, using a dominance weighting of the RS. This method has been exploited in the context of melody extraction from polyphonic music signals.

The work presented in this thesis is aligned with a renewed interest in design for opening new possibilities for democracy. This thesis builds on a growing corpus of research investigating the role of design in supporting the formation of publics. In this thesis, the concept of publics is aligned with Dewey’s view, which refers to heterogeneous groups of people concerned about an issue who organize themselves to address it. This thesis aims to contribute to this corpus of research by investigating the following research questions: what design processes can contribute to the formation of publics? and what design interventions can enable these processes? Answers to these questions are constructed by engaging in a practice-based design research of a case study of dyslexia in Trentino, a region in Italy where dyslexia is a controversial issue grounded not only in medical but also societal and political conditions. The main contribution of the thesis is a method to support the formation of publics following a practice-based interaction design approach. This method proposes articulating, representing and reconfiguring as three intertwining and complementing processes that can support the formation of publics. In addition, it proposes that designing interventions on the basis of physical artefacts, digital platforms and events can enable people to act on an issue. Finally, it proposes programs for action as takeaways of design research projects that aim to enable people to act on societal issues.

Cities are essential crucibles for innovation, novelty, economic prosperity and diversity. They are not a mere reflection of individual characteristics, but instead the result of a complex interaction between people and space. Yet, little is known about this self-organized and complex relationship. Traditional approaches have either used surveys to explain in detail how a few individuals experience bits of a city, or considered cities as a whole from their outputs (e.g. total crimes). This tide has however tuned in recent years: the availability of new sources of data have allowed to observe, describe, and predict human behaviour in cities at an unprecedented scale and detail. This thesis adopts a "data mining approach" where we study urban spaces combining new sources of automatically collected data and novel statistical methods. Particularly, we focus on the relationship between the built environment, described by census information, geographical data, and images, and human behaviour proxied by extracted from mobile phone traces. The contribution of our thesis is two-fold. First, we present novel methods to describe urban vitality, by collecting and combining heterogeneous data sources. Second, we show that, by studying the built environment in conjunction with human behaviour, we can reliably estimate the effect of neighbourhood characteristics, predict housing prices and crime. Our results are relevant to researchers within a broad range of fields, from computer science to urban-planning and criminology, as well as to policymakers. The thesis is structured in two parts. In the first part, we investigate what creates urban life. We operationalize the theory of Jane Jacobs, one of the most famous authors in urban planning, who suggested that the built environment and vitality are intimately connected. Using Web and Open data to describe neighbourhoods, and mobile phone records to proxy urban vitality, we show that it is possible to predict vitality from the built environment, thus confirming Jacob's theory. Also, we investigate the effect of safety perception on urban vitality by introducing a novel Deep Learning model that relies on street-view images to extract security perception. Our model describes how perception modulates the relative presence of females, elderly and younger people in urban spaces. Altogether, we demonstrate how objective and subjective characteristics describe urban life. In the second part of this dissertation, we outline two studies that stress the importance of considering, at the same time, multiple factors to describe cities. First, we build a predictive model showing that objective and subjective neighbourhood features drive more than 20% of the housing price. Second, we describe the effect played by a neighbourhood's characteristics on the presence of crime. We present a Bayesian method to compare two alternative criminology theory, and show that the best description is achieved by considering together socio-economic characteristics, built-environment, and mobility of people. Also, we highlight the limitations of transferring what we learn from one city to another. Our findings show that new sources of data, automatically sensed from the environment, can complement the lengthy and costly survey-based collection of urban data and reliably describe neighbourhoods at an unprecedented scale and breath. We anticipate that our results will be of interest to researchers in computer science, urban planning, sociology and more broadly, computational social science.

Most human language understanding is grounded in perception. There is thus growing interest in combining information from language and vision. Multiple models based on Neural Networks have been proposed to merge language and vision information. All the models share a common backbone consisting of an encoder which learns to merge the two types of representation to perform a specific task. While some models have seemed extremely successful on those tasks, it remains unclear how the reported results should be interpreted and what those models are actually learning. Our contribution is three-fold. We have proposed (a) a new model of Visually Grounded Dialogue; (b) a diagnostic dataset to evaluate the encoder ability to merge visual and language input; (c) a method to evaluate the quality of the multimodal representation computed by the encoder as general purposed representations. We have proposed and analyzed a cognitive plausible architecture in which dialogue system modules are connected through a common \emph{grounded dialogue state encoder}. Our in-depth analysis of the dialogues shows the importance of going beyond task-success in the evaluation of Visual Dialogues: the dialogues themselves should play a crucial role in such evaluation. We have proposed a diagnostic dataset, \emph{FOIL} which consists of images associated with incorrect captions that the model has to detect and correct. Finally, we have used FOIL to evaluate the quality of the multimodal representation produced by an encoder trained on different multimodal tasks. We have shown how the training task used effects the stability of the representation, their transferability and the model confidence.

Exploratory search is the new frontier of information consumption as it goes well beyond simple \emph{lookups}. Information repositories are ubiquitous and grow larger every day, and automated search systems help users find information in such collections. To extract knowledge from these repositories, the common ``query lookup'' retrieval paradigm accepts a set of specifications (the query) that describes the objects of interest and then collects such objects. Yet, the query lookup retrieval paradigms commonly in use are no more sufficient to support complex information needs, as they can only provide candidate starting points, but do not help the user in expanding their knowledge. To ease access and consumption of rich information repositories, we address the crucial problem of data exploration. Exploratory tasks match the natural need for finding answers to open-ended information needs within an unfamiliar environment. In particular, in this dissertation, we focus on enabling access to and exploration of rich information graphs. Within businesses, organizations, and among researchers, data is produced in many forms, large volumes, and different contexts. As a consequence of this heterogeneity, many applications find more useful modelling their datasets with the graph model, where information is represented with entities (nodes) and relationships (edges). Those are the data graphs, the graph databases, the knowledge graphs, or more generally information graphs. The richness of their schema and of their content makes it challenging for users to express appropriate queries and retrieve the desired results. Hence, to allow an effective exploration of a graph, we require: (i) an expressive \emph{query paradigm}, (ii) an intuitive \emph{query mechanism}, and (iii) an appropriate \emph{storage and query processing system}. In this work, we address these three requirements. An exploratory query should be simple enough to avoid complicate declarative languages (such as SQL or SPARQL), and at the same time, it should retain the flexibility and expressiveness of such languages. For this reason, with respect to the query paradigm, we introduce the notion of \emph{exemplar queries} and propose extensions to handle multiple incomplete examples. An exemplar query is a query method in which the user, or the analyst, circumvents query languages by using examples as input. In particular, the solution we design allows flexible matching in the case of incomplete or partially specified examples. Moreover, to enable this query paradigm, there is the need for interactive systems that implement an incremental query-constructions mechanism and interactive explorations. To address this need, we study algorithms and implementations based on pseudo-relevance feedback for \emph{exemplar query suggestion}, along with an in-depth study of their effectiveness. Finally, as there exist many graph databases, high heterogeneity can be observed in the functionalities and performances of these systems. We provide an exhaustive evaluation methodology and a comprehensive study of the existing systems that allow to understand their capabilities and limitations. In particular, we design a novel micro-benchmarking framework for the assessment of the functionalities of some graph databases among the most prominent in the area and provide detailed insights on their performance.

A key challenge posed by the Next Generation Internet landscape, is that modern service-based systems need to cope with open and continuously evolving environments and to operate under dynamic circumstances. Dynamism is given by changes in the operational context, changes in the availability of resources and variations in their behavior, changes in users goals, etc. Indeed, dynamically discover, select and compose the appropriate services in open and expanding domains is a challenging task. Many approaches for self-adaptive systems have been proposed in the last decades. Unfortunately, although they support run-time adaptation, current approaches tend to foresee the system adaptation requirements and their related solutions at design-time. This makes them inadequate for the application in open environments, where components constantly join/leave the system, since they require for continuous involvement of IT and domain experts for the systems re-configuration. We claim that a new way of approaching the adaptation of systems is needed. In this dissertation, we propose a novel design for adaptation framework for modeling and executing modern service-based systems. The idea of the approach consists in defining the complete life-cycle for the continuous development and deployment of service-based systems, by facilitating (i) the continuous integration of new services that can easily join the systems, and (ii) the systems operation under dynamic circumstances, to face the openness and dynamicity of the environment. Furthermore, Collective Adaptive Systems (CAS) are spreading in new emerging contexts, such as the shared economy trend. Modern systems are expected to handle a multitude of heterogeneous components that cooperate to accomplish collective tasks. In these settings, an extension of our framework in the direction of CAS has also been defined. The core enablers of the proposed framework have been implemented and evaluated in real-world scenarios in the mobility domain. Promising evaluation results demonstrate their practical applicability.

This thesis aims to address a significant issue related with the consumption of diversified data in the field of semantics and knowledge representation by using a framework which allows the data consumption in a generic, scalable and pleasing manner. The work proposes a mixed solution by splitting the issue into four subproblems: how to preserve the richness associated with the data; how to present information about an object in a single or multiple visualization contexts; and to provide a seamless exploration of interconnected entities; and how to design a tool that offers a better user experience. A real-world object can have various representations which lead to data diversity. However, each representation captures a view (mostly partial) of an object. To preserve the richness associated with the data, we follow an entity-centric design approach. In this approach, we represent multiple datasets related to an object as an entity with various properties. An entity is then further categorized in a group according to its similarities or differences. Our contextual model not only considers the transformation of objects as entities but also adapts to various visualization contexts. These contexts are space, list, timeline, and network. We design a multiview visualization framework that allows simultaneous presentation of entities according to these four defined visualization contexts. To allow seamless interaction of data with the users, we emphasized on using a multilayered architecture where: 1)datasets are aggregated and stored using an entity-centric approach, 2) visualized in various contexts and viewpoints simultaneously according to the entity types and users' need. This adaptation is capable enough to facilitate presentation and exploration of diversified data according to users need. To prove the feasibility of our framework, we applied it to visualize diversified data in various settings. Continuous interaction with the end users produced valuable feedback and essential design suggestions. Finally, multiple prototypes were evaluated with the end users to verify their usability. The results obtained were highly favorable.

In this thesis, we address the design space exploration of cyber-physical system architectures to select correct-by-construction configuration and interconnection of system components taken from pre-defined libraries. We formulate the exploration problem as a mapping problem and use optimization to solve it by searching for a minimum cost architecture that meets system requirements. Using a graph-based representation of a system architecture, we define a set of generic mixed integer linear constraints over graph vertices, edges and paths, and use these constraints to instantiate a variety of design requirements (e.g., interconnection, flow, workload, timing, reliability, routing). We implement a comprehensive toolbox that supports all steps of the proposed methodology. It provides a pattern-based formal language to facilitate requirements specification and a set of scalable algorithms for encoding and solving exploration problems. We prove our concepts on a set of case studies for different cyber-physical system domains, such as electrical power distribution networks, reconfigurable industrial production lines and wireless sensor networks.