Search

Multiple tasks have been used to demonstrate the relation between numbers and space. The classic interpretation of these directional spatial–numerical associations (d-SNAs) is that they are the product of a mental number line (MNL), in which numerical magnitude is intrinsically associated with spatial position. The alternative account is that d-SNAs reflect task demands, such as explicit numerical judgements and/or categorical responses. In the novel “Where was The Number?” task, no explicit numerical judgements were made. Participants were simply required to reproduce the location of a numeral within a rectangular space. Using a between-subject design, we found that numbers, but not letters, biased participants’ responses along the horizontal dimension, such that larger numbers were placed more rightward than smaller numbers, even when participants completed a concurrent verbal working memory task. These findings are consistent with the MNL account, such that numbers specifically are inherently left-to-right oriented in Western participants.

Shape estimation of soft robotic systems is challenging due to the range of deformations that can be achieved, and the limited availability of physically compatible sensors. We propose a method of reconstruction using Inertial Measurement Units (IMUs), which are mounted on segments of a deformable manipulator. This approach utilizes the piecewise constant curvature model in combination with the quaternion data from IMUs to allow for accuracy reconstruction and closed-loop control. A key strength of this approach is that it is hardware agnostic, and could be used on any soft structure to provide pose reconstruction and controllability. We explore this approach experimentally on a growing, extendable 3D printed continuum body structure, demonstrating that high accuracy reconstruction that can be achieved.

The more people commute by bicycle, the higher is the number of cyclists using their smartphones while cycling and compromising traffic safety. We have designed, implemented and evaluated two prototypes for smartphone control devices that do not require the cyclists to remove their hands from the handlebars - the three-button device Tribike and the rotation-controlled Brotate. The devices were the result of a user-centred design process where we identified the key features needed for a on-bike smartphone control device. We evaluated the devices in a biking exercise with 19 participants, where users completed a series of common smartphone tasks. The study showed that Brotate allowed for significantly more lateral control of the bicycle and both devices reduced the cognitive load required to use the smartphone. Our work contributes insights into designing interfaces for cycling., 22nd International Conference on Human-Computer Interaction with Mobile Devices and Services (MobileHCI '20), October 5--8, 2020, Oldenburg, Germany

In this study, we attempt to specify the cognitive support behavior of a previously designed embodied conversational agent coach that provides learning support to low-literates. Three knowledge gaps are identified in the existing work: an incomplete specification of the behaviors that make up ‘support,’ an incomplete specification of how this support can be personalized, and unclear speech recognition rules. We use the socio-cognitive engineering method to update our foundation of knowledge with new online banking exercises, low-level scaffolding and user modeling theory, and speech recognition. We then refine the design of our coach agent by creating comprehensive cognitive support rules that adapt support based on learner needs (the ‘Generalized’ approach) and attune the coach’s support delay to user performance in previous exercises (the ‘Individualized’ approach). A prototype is evaluated in a 3-week within- and between-subjects experiment. Results show that the specified cognitive support is effective: Learners complete all exercises, interact meaningfully with the coach, and improve their online banking self-efficacy. Counter to hypotheses, the Individualized approach does not improve on the Generalized approach. Whether this indicates suboptimal operationalization or a deeper problem with the Individualized approach remains as future work.

Shape perception is crucial for object recognition. However, it remains unknown exactly how shape information is represented, and, consequently, used by the visual system. Here, we hypothesized that the visual system represents “shape skeletons” to both (1) perceptually organize contours and component parts into a shape percept, and (2) compare shapes to recognize objects. Using functional magnetic resonance imaging (fMRI) and representational similarity analysis (RSA), we found that a model of skeletal similarity explained significant unique variance in the response profiles of V3 and LO, regions known to be involved in perceptual organization and object recognition, respectively. Moreover, the skeletal model remained predictive in these regions even when controlling for other models of visual similarity that approximate low- to high-level visual features (i.e., Gabor-jet, GIST, HMAX, and AlexNet), and across different surface forms, a manipulation that altered object contours while preserving the underlying skeleton. Together, these findings shed light on the functional roles of shape skeletons in human vision, as well as the computational properties of V3 and LO.

Categorization of everyday objects requires that humans form representations of shape that are tolerant to variations among exemplars. Yet, how such invariant shape representations develop remain poorly understood. By comparing human infants (6-12 months; N = 82) to computational models of vision using comparable procedures, we shed light on the origins and mechanisms underlying object perception. Following habituation to a never-before-seen object, infants classified other novel objects across variations in their component parts. Comparisons to several computational models of vision, including models of high- and low-level vision, revealed that infants’ performance was best described by a model of shape based on skeletal structure. Interestingly, infants outperformed a range of artificial neural networks (ANNs), selected for their massive object experience and biological plausibility, under the same conditions. Altogether, these findings suggest that robust representations of shape can be formed with little language or object experience by relying on perceptually invariant skeletal structure.

A representation of shape that is low dimensional and stable across minor disruptions is critical for object recognition. Computer vision research suggests that such a representation can be supported by the medial axis-a computational model for extracting a shape's internal skeleton. However, few studies have shown evidence of medial axis processing in humans, and even fewer have examined how the medial axis is extracted in the presence of disruptive contours. Here, we tested whether human skeletal representations of shape reflect the medial axis transform (MAT), a computation sensitive to all available contours, or a pruned medial axis, which ignores contours that may be considered ''noise.'' Across three experiments, participants (N = 2062) were shown complete, perturbed, or illusory two-dimensional shapes on a tablet computer and were asked to tap the shapes anywhere once. When directly compared with another viable model of shape perception (based on principal axes), participants' collective responses were better fit by the medial axis, and a direct test of boundary avoidance suggested that this result was not likely because of a task-specific cognitive strategy (Experiment 1). Moreover, participants' responses reflected a pruned computation in shapes with small or large internal or external perturbations (Experiment 2) and under conditions of illusory contours (Experiment 3). These findings extend previous work by suggesting that humans extract a relatively stable medial axis of shapes. A relatively stable skeletal representation, reflected by a pruned model, may be well equipped to support real-world shape perception and object recognition.

With seemingly little effort, humans can both identify an object across large changes in orientation and extend category membership to novel exemplars. Although researchers argue that object shape is crucial in these cases, there are open questions as to how shape is represented for object recognition. Here we tested whether the human visual system incorporates a three-dimensional skeletal descriptor of shape to determine an object’s identity. Skeletal models not only provide a compact description of an object’s global shape structure, but also provide a quantitative metric by which to compare the visual similarity between shapes. Our results showed that a model of skeletal similarity explained the greatest amount of variance in participants’ object dissimilarity judgments when compared with other computational models of visual similarity (Experiment 1). Moreover, parametric changes to an object’s skeleton led to proportional changes in perceived similarity, even when controlling for another model of structure (Experiment 2). Importantly, participants preferentially categorized objects by their skeletons across changes to local shape contours and non-accidental properties (Experiment 3). Our findings highlight the importance of skeletal structure in vision, not only as a shape descriptor, but also as a diagnostic cue of object identity.

This workshop discusses the balance between safety and productivity as automated vehicles turn into 'mobile offices': spaces where non-driving activities are performed during one's daily commute. Technological developments reduce the active role of the human driver that might, nonetheless, require occasional intervention. To what extent are drivers allowed to dedicate resources to non-driving work-related activities? To address this critical question, the workshop brings together a diverse community of researchers and practitioners that are interested in questions as follows: what non-driving activities are likely to be performed on one's way to work and back; what is a useful taxonomy of these tasks; how can various tasks be studied in experimental settings; and, what are the criteria to assess human performance in automated vehicles. To foster further dialogue, the outcome of the workshop will be an online blog where attendees can contribute their own thoughts: https://medium.com/the-mobile-office.

Augmented reality technology is a unique medium that can be helpful for young children's entertainment and education, but in order to achieve the benefits of this technology, augmented reality experiences need to be appropriately designed for young children's developing physical and cognitive skills. In the present study we investigated how 5-10 year-old children react to typical handheld augmented reality interaction techniques such as crosshair selection and finger selection, in AR environments that require them to change perspective or not. Our analysis shows significant impacts of age upon AR performance, with young children having slower selection times, more tracking losses, and taking longer to recover tracking. Significant differences were also found between AR interaction technique conditions, with finger selection being faster than crosshair selection, and interactions which required changes in perspective taking longer, generating more tracking losses, and more errors in selection. Furthermore, by analyzing children's performance in relation to metrics of physical and cognitive development, we identified correlations between AR interaction techniques performance and developmental tests of spatial relations, block construction and visuomotor precision. Gender differences were analyzed but no significant effects were detected.

The study of spatial cognition has benefited greatly from a technique known as the disorientation procedure. This procedure was originally used with rats to show that they relied on the geometry of an enclosed space to locate a target hidden in that space. Disorientation has since been used with a variety of mobile animals, including human children, to examine the coding of geometric information. Here, we focus mostly on our recent work with young children. We examine a set of issues concerning reorientation--namely, the nature of geometric coding, the processes invoked by disorientation, and the developmental origins of using geometric information to determine location. We have employed a variety of methods to examine these issues; the methods include analyzing search behaviors, using spaces of different shapes, varying viewing position, and comparing different disorientation procedures. The implications for how children and nonhuman animals code geometric information are discussed.

Spatial reasoning and visualization skills are critical skills for Computer Science, Engineering, and Technology (CSET) students. The Purdue Spatial Visualization Test: Visualizations of Rotations (PSVT:R) is considered the "gold standard" of spatial assessments and is a paper-and-pencil test that requires the mental rotation of three-dimensional connected cubes. Consistently, results on this test show sex differences, with men performing better than women. Spatial cognition is more than just mental rotation, however, and CSET education needs a variety of assessment tools and methods. In this session, participants will engage in a variety of spatial reasoning and visualization problems to expand their understanding of this area. We will explore different ways to assess spatial skills and consider the complex interactions of context, resources, and multiple strategies. By reviewing the literature on how children develop spatial understanding, CSET faculty can expand the options they consider when designing learning opportunities and assessment methods for college students. Another thread woven throughout this session is the consideration of differential performance of subgroups on various tests of spatial skills. The PSVT:R and tests like it often show sex differences and SES differences in performance, especially under timed-response conditions. In this session, we will consider potential explanations for these differences and outline some open areas for research.

The goal was to identify the contribution of intrinsic mechanical properties of the muscular-skeletal system and the reflex gains in controlling balance during standing. The combination of balance perturbations experiments and closed loop identification schemes made it possible to identify the reflex loop gains and intrinsic mechanical properties in various environmental conditions. Human balance responses were studied by placing subjects on a movable support base while keeping their eyes open or closed. EMG, body motion and the ground reaction forces were recorded. From the platform perturbation data the frequency response functions of the controller dynamics and the admittance of the balance control system were estimated with a non-parametric closed loop identification technique. Using a parametric model of balance control and a fitting procedure the sum of the intrinsic stiffness and neural position feedback gain, the neural time delay, the neural velocity feedback gain and the intrinsic damping were uniquely identified. The results show that subjects balancing on a randomly moving platform in the forward-backward direction applied a minimal stiffness strategy. The required damping to avoid oscillations was mainly due to neural velocity feedback rather than to intrinsic damping.

The proposal of Rips et al. is motivated by discontinuity and input claims. The discontinuity claim is that no continuity exists between early (nonverbal) numerical representations and natural number. The input claim is that particular experiences (e.g., cardinality-related talk and object-based activities) do not aid in natural number construction. We discuss reasons to doubt both claims in their strongest forms.

Dynamic Bayesian networks have been well explored in the literature as discrete-time models: however, their continuous-time extensions have seen comparatively little attention. In this paper, we propose the first implementation of a constraint-based algorithm for learning the structure of continuous-time Bayesian networks. We discuss the different statistical tests and the underlying hypotheses used by our proposal to establish conditional independence. Furthermore, we analyze and discuss the computational complexity of the best and worst cases for the proposed algorithm. Finally, we validate its performance using synthetic data, and we discuss its strengths and limitations comparing it with the score-based structure learning algorithm from Nodelman et al. [23] . We find the latter to be more accurate in learning networks with binary variables, while our constraint-based approach is more accurate with variables assuming more than two values. Numerical experiments confirm that score-based and constraint-based algorithms are comparable in terms of computation time.

Background:Recently, mobile devices, such as smartphones, have been introduced into healthcare research to substitute paper diaries as data-collection tools in the home environment. Such devices support collecting patient data at different time points over a long period, resulting in clinical time-series data with high temporal complexity, such as time irregularities. Analysis of such time series poses new challenges for machine-learning techniques. The clinical context for the research discussed in this paper is home monitoring in chronic obstructive pulmonary disease (COPD).Objective:The goal of the present research is to find out which properties of temporal Bayesian network models allow to cope best with irregularly spaced multivariate clinical time-series data.Methods:Two mainstream temporal Bayesian network models of multivariate clinical time series are studied: dynamic Bayesian networks, where the system is described as a snapshot at discrete time points, and continuous time Bayesian networks, where transitions between states are modeled in continuous time. Their capability of learning from clinical time series that vary in nature are extensively studied. In order to compare the two temporal Bayesian network types for regularly and irregularly spaced time-series data, three typical ways of observing time-series data were investigated: (1) regularly spaced in time with a fixed rate; (2) irregularly spaced and missing completely at random at discrete time points; (3) irregularly spaced and missing at random at discrete time points. In addition, similar experiments were carried out using real-world COPD patient data where observations are unevenly spaced.Results:For regularly spaced time series, the dynamic Bayesian network models outperform the continuous time Bayesian networks. Similarly, if the data is missing completely at random, discrete-time models outperform continuous time models in most situations. For more realistic settings where data is not missing completely at random, the situation is more complicated. In simulation experiments, both models perform similarly if there is strong prior knowledge available about the missing data distribution. Otherwise, continuous time Bayesian networks perform better. In experiments with unevenly spaced real-world data, we surprisingly found that a dynamic Bayesian network where time is ignored performs similar to a continuous time Bayesian network.Conclusion:The results confirm conventional wisdom that discrete-time Bayesian networks are appropriate when learning from regularly spaced clinical time series. Similarly, we found that time series where the missingness occurs completely at random, dynamic Bayesian networks are an appropriate choice. However, for complex clinical time-series data that motivated this research, the continuous-time models are at least competitive and sometimes better than their discrete-time counterparts. Furthermore, continuous-time models provide additional benefits of being able to provide more fine-grained predictions than discrete-time models, which will be of practical relevance in clinical applications.

This paper describes a decision support system to fulfill the diagnosis of post-harvest diseases in apple fruit. The diagnostic system builds on user feedback elicitation via multiple conversational rounds. The interaction with the user is conducted by selecting and displaying images depicting symptoms of diseased apples. The system is able to adapt the image selection mechanism by exploiting previous user feedback through a contextual multi-armed bandit approach. We performed a large scale user experiment, where different strategies for image selection have been compared in order to identify which reloading strategy makes users more effective in the diagnosis task. Concretely, we compared contextual multi-armed bandit methods with two baseline strategies and identified that the exploration-exploitation principle significantly paid off in comparison to a greedy and to a random stratified selection strategy. Although the application context is very domain specific, we believe that the general methodology of conversational item selection for eliciting user preferences applies to other scenarios in decision support and recommendation systems.

Emerging studies in the deep learning community focus on techniques aimed to identify which part of a graph can be suitable for making better decisions and best contributes to an accurate inference. These researches (i.e., “attentional mechanisms” for graphs) can be applied effectively in all those situations in which it is not trivial to capture dependency between the involved entities while discharging useless information. This is the case, e.g., of functional connectivity in human brain, where rapid physiological changes, artifacts and high inter-subject variability usually require highly trained clinical expertise. In order to evaluate the effectiveness of the attentional mechanism in such critical situation, we consider the task of normal vs abnormal EEG classification using brain network representation of the corresponding EEG recorded signals.

The current understanding of deep neural networks can only partially explain how input structure, network parameters and optimization algorithms jointly contribute to achieve the strong generalization power that is typically observed in many real-world applications. In order to improve the comprehension and interpretability of deep neural networks, we here introduce a novel theoretical framework based on the compositional structure of piecewise linear activation functions. By defining a direct acyclic graph representing the composition of activation patterns through the network layers, it is possible to characterize the in-stances of the input data with respect to both the predicted label and the specific (linear) transformation used to perform predictions. Preliminary tests on the MNIST dataset show that our method can group input instances with regard to their similarity in the internal representation of the neural network, providing an intuitive measure of input complexity.

Incomplete data are a common feature in many domains, from clinical trials to industrial applications. Bayesian networks (BNs) are often used in these domains because of their graphical and causal interpretations. BN parameter learning from incomplete data is usually implemented with the Expectation-Maximisation algorithm (EM), which computes the relevant sufficient statistics (&ldquo, soft EM&rdquo, ) using belief propagation. Similarly, the Structural Expectation-Maximisation algorithm (Structural EM) learns the network structure of the BN from those sufficient statistics using algorithms designed for complete data. However, practical implementations of parameter and structure learning often impute missing data (&ldquo, hard EM&rdquo, ) to compute sufficient statistics instead of using belief propagation, for both ease of implementation and computational speed. In this paper, we investigate the question: what is the impact of using imputation instead of belief propagation on the quality of the resulting BNs? From a simulation study using synthetic data and reference BNs, we find that it is possible to recommend one approach over the other in several scenarios based on the characteristics of the data. We then use this information to build a simple decision tree to guide practitioners in choosing the EM algorithm best suited to their problem.

A necessary step in the development of artificial intelligence is to enable a machine to represent how the world works, building an internal structure from data. This structure should hold a good trade-off between expressive power and querying efficiency. Bayesian networks have proven to be an effective and versatile tool for the task at hand. They have been applied to modeling knowledge in a variety of fields, ranging from bioinformatics to law, from image processing to economic risk analysis. A crucial aspect is learning the dependency graph of a Bayesian network from data. This task, called structure learning, is NP-hard and is the subject of intense, cutting-edge research. In short, it can be thought of as choosing one graph over the many candidates, grounding our reasoning over a collection of samples of the distribution generating the data. The number of possible graphs increases very quickly at the increase in the number of variables. Searching in this space, and selecting a graph over the others, becomes quickly burdensome. In this survey, we review the most relevant structure learning algorithms that have been proposed in the literature. We classify them according to the approach they follow for solving the problem and we also show alternatives for handling missing data and continuous variable. An extensive review of existing software tools is also given.

This chapter builds on a dataset where online users of a spa platform participated in an online idea contest providing free-text descriptions of their proposals for spa services. A panel of domain experts annotated these idea descriptions with a score for their innovativeness that serves as ground truth for machine learning experiments. Thus, the contribution lies in the application of topic modeling techniques to free-text idea descriptions in order to automatically identify innovative proposals based on advanced text processing and machine learning. Results of this case study indicate that topic modeling can outperform the ZeroR baseline as well as traditional survey scales for lead user identification and therefore constitute a first step towards exploring this technique for innovation research.

This demo presents a conversational navigation approach for a diagnostic application of postharvest diseases of apple with the goal to educate users on the diagnosed diseases as well as to recommend consequences for the storage facility and what action to take for the next growing period. It thus builds on earlier works on picture-based navigation for conversational recommender systems and provides evidence for its usability based on a first small-scale comparative usability study. © 2018 Copyright held by the owner/author(s)

Non-stationary continuous time Bayesian networks are introduced. They allow the parents set of each node to change over continuous time. Three settings are developed for learning non-stationary continuous time Bayesian networks from data: known transition times, known number of epochs and unknown number of epochs. A score function for each setting is derived and the corresponding learning algorithm is developed. A set of numerical experiments on synthetic data is used to compare the effectiveness of non-stationary continuous time Bayesian networks to that of non-stationary dynamic Bayesian networks. Furthermore, the performance achieved by non-stationary continuous time Bayesian networks is compared to that achieved by state-of-the-art algorithms on four real-world datasets, namely drosophila, saccharomyces cerevisiae, songbird and macroeconomics.

Continuous time Bayesian networks offer a compact representation for modeling structured stochastic processes that evolve over continuous time. In these models, the time duration that a variable stays in a state until a transition occurs is assumed to be exponentially distributed. In real-world scenarios, however, this assumption is rarely satisfied, in particular when describing more complex temporal processes. To relax this assumption, we propose an extension to support the modeling of the transitioning time as a hypoexponential distribution by introducing an additional hidden variable. Using such an approach, we also allow CTBNs to obtain memory, which is lacking in standard CTBNs. The parameter estimation in the proposed models is transformed into a learning task in their equivalent Markovian models.

The deep learning wave is propagating through many research areas and communities. In the last years it quickly propagated to Recommendation Systems, a research area which aims to recommend items to users. Indeed, many deep learning models and architectures have been proposed for Recommendation Systems to improve collaborative filtering and content based algorithms. In this paper we propose a hybrid recommendation system combining user ratings and natural language text processing to solve the 0/1 recommendation problem. In particular, we describe a deep learning architecture combining two information sources, namely natural language text and user rating. Natural language text is used to learn a user-specific content-based classifier, while user ratings are used to develop user-adaptive collaborative filtering recommendations. We perform numerical experiments on MovieLens 1M and reach first preliminary, but promising results, showing the proposed architecture has the potential to combine content-based and collaborative filtering recommendation mechanisms using a deep learning supervisor.

Classification and clustering of streaming data are relevant in finance, computer science, and engineering while they are becoming increasingly important in medicine and biology. Streaming data are analyzed with algorithms and models capable to represent dynamics, sequences and time. Dynamic Bayesian networks and hidden Markov models are commonly used to analyze streaming data. However, they are concerned with evenly spaced time series data and thus suffer from several limitations. Indeed, it is not clear how timestamps should be discretized even if some approaches to mitigate this problem have been recently made available. In this paper we describe the class of continuous time Bayesian networks classifiers and develop algorithms for their parametric and structural learning to solve classification and clustering of multivariate discrete state continuous time trajectories. Numerical experiments on synthetic and real world data are used to compare the performance of continuous time Bayesian network models to that achieved by dynamic Bayesian networks. In particular, post-stroke rehabilitation data is used for the classification task while urban traffic data from continuous time loop is used for the clusteirng task. The achieved results confirm the effectiveness of the proposed approaches.

There has been in the past years a widespread effort in the computer science community to extend the vast information existing on the Internet with “semantic structure” , in the form of ontologies and their associated inference systems, that can be exploited to support complex processes and services as can be carried out by dedicated software agents, as well as the structured querying of information as in database management systems. The Semantic Web [2] and its evolution into the Web of Data [3] represent the most noticeable contributions to this trend. At the core of these approaches there is the idea of replacing the simple Web publishing that has characterized the growth of the WWW since its inception with a more sophisticated form of publishing where, beside the normal content, ontologies and other semantic information is added as meta-content. The theoretical merits of this enterprise notwithstanding, its practical implementation still meets considerable hurdles as identified, for instance, by Halevy et al. [4] : namely, cost of ontology writing and maintenance, technical challenges of Semantic Web publishing, eventual unwillingness of the communities involved in the evolution of an ontology to cooperate to such an evolution, lack of resilience of standard inference systems in the coping with the possibility of mistaken premises and of incorrect data. Our purpose here is to fulfill the objective of a Web-based information space that becomes more usable and meaningful as a consequence of making explicit its underlying semantic structure, but we do so by reversing the direction taken by such frameworks as the Semantic Web and the Web of Data. Indeed, we stick to the general criterion, proposed and argued by Halevi et al. [4] , according to which semantic interpretation, based on techniques derived from artificial intelligence and statistical inference, is easy compared to the daunting task of semantic publishing. Thus, we make a kind of semantic structure dynamically emerge from the available unstructured information, as opposed to superimpose it in a pre-defined format. In this we follow also the lines of the program described in Arcelli Fontana et al. [1], which aims at generating and discovering concepts on content-based networks, such as the Web, by abstracting and integrating information existing within independent units (such as Web pages, documents and other kinds of textual objects) distributed across a content network. The main tools that we use in order to implement our approach are taken from the area of Probabiilistic Topic Modeling [6]. Techniques from this area provide a sound background for developing an effective methodology for extracting concepts in the form of topics from existing information networks and then connecting them into new networks that highlight the semantic structure and relationships of the thus interpreted information. Indeed, the two main contributions of our methodology are in the construction of two networks, that are overlaid one on the other: - A higher-level topic-topic network where topics are related in terms of their semantic proximity; - A finer-grained object-object network where the textual objects that make up the topics and account for the relationships among them are analyzed in terms of their specific relationships, with links that trespass topic boundaries, again using semantic proximity as a criterion for establishing object-object links. This methodology has been tested through “knowledge sets” that have been obtained from the public YAGO2 knowledge base with the following approach: starting from specific thematic sectors of the knowledge base, we have created a number of corresponding knowledge kernels in terms of the Wikipedia pages associated with the chosen sectors, and we have then expanded such kernels by following their Web hyperlinks until suitable sets for experimentation have been created. In this way we have been able to carry out experiments, that will be here described, on knowledge sets for “Green Economy”, “Touchscreen Mobile Phones”, “City Cars” and “Terrorism”. From the practical point of view of exploring and analyzing the various knowledge sets, this has had the result of highlighting, for instance, the diverse relationships between the domain of terrorism with the domain of public transportation networks, where many terroristic attempts have been, more or less successfully, carried out. Clearly, aside from the their conceptual clustering through the creation of topic-topic networks, the effect of the creation of object-object networks on the Web pages that constitute the various knowledge sets is of connecting through "semantic links" Web pages that were previously unlinked through the existing hyperlinks. Given that adding links on a semantic basis to formerly unlinked Web objects is also one of the goals of the Web of Data, we show how our approach might contribute for all practical purposes to attaining this goal, even if through different methodological and technical premises. We also highlight differences and relationships with work on community detection in complex networks and systems, as exemplified by the well-known Newman-Girvan algorithm [5].

Prediction of foreign exchange (FX) rates is addressed as a binary classification problem in which a continuous time Bayesian network classifier (CTBNC) is developed and used to solve it. An exact algorithm for inference on CTBNC is introduced. The performance of an instance of these classifiers is analysed and compared to that of dynamic Bayesian network by using real tick by tick FX rates. Performance analysis and comparison, based on different metrics such as accuracy, precision, recall and Brier score, evince a predictive power of these models for FX rates at high frequencies. The achieved results also show that the proposed CTBNC is more effective and more efficient than dynamic Bayesian network classifier. In particular, it allows to perform high frequency prediction of FX rates in cases where dynamic Bayesian networks-based models are computationally intractable.

When looking for recently published scientific papers, a researcher usually focuses on the topics related to her/his scientific interests. The task of a recommender system is to provide a list of unseen papers that match these topics. The core idea of this paper is to leverage the latent topics of interest in the publications of the researchers, and to take advantage of the social structure of the researchers (relations among researchers in the same field) as reliable sources of knowledge to improve the recommendation effectiveness. In particular, we introduce a hybrid approach to the task of scientific papers recommendation, which combines content analysis based on probabilistic topic modeling and ideas from collaborative filtering based on a relevance-based language model. We conducted an experimental study on DBLP, which demonstrates that our approach is promising.

Most evaluations of novel algorithmic contributions assess their accuracy in predicting what was withheld in an of- ine evaluation scenario. However, several doubts have been raised that standard offline evaluation practices are not ap- propriate to select the best algorithm for field deployment. The goal of this work is therefore to compare the offline and the online evaluation methodology with the same study participants, i.e. a within users experimental design. This paper presents empirical evidence that the ranking of algo- rithms based on offline accuracy measurements clearly con- tradicts the results from the online study with the same set of users. Thus the external validity of the most commonly applied evaluation methodology is not guaranteed.

Physical activity (PA) is considered one of the most important factors for the prevention and management of non-communicable diseases (NCDs). Mobile technologies offer several opportunities for supporting PA, especially if combined with psychological aspects, model-based reasoning systems and personalized human computer interaction. This still on-going research aims at developing a scalable framework that targets PA promotion among both clinical and non-clinical population, exploiting Bayesian Networks and Expert Systems to characterize and predict qualitative variables like self-efficacy. The expected outcomes are the collection and management of real-time behavioral and psychological data to define a personalized strategy for increasing PA.

User generated content in general and textual reviews in particular constitute a vast source of information for the decision making of tourists and management and are therefore a key component for e-tourism. This paper provides a description of the topic model method with a particular application focus on the tourism domain. It therefore contributes different application scenarios where the topic model method processes textual reviews in order to provide decision support and recommendations to online tourists as well as to build a basis for further analytics. In the latter case the delivery of additional semantics helps digging into the enormous amounts of data that are continuously collected in present time. The contribution therefore consists of new models based on the topic model method and results from experimenting with user generated review data on restaurants and hotels.

We apply machine learning to detect design patterns in software systems.We exploit a specific design pattern model to apply machine learning techniques.We compare the performances of several machine learning algorithms.We provide a large dataset containing manually checked design pattern instances. The detection of design patterns is a useful activity giving support to the comprehension and maintenance of software systems. Many approaches and tools have been proposed in the literature providing different results. In this paper, we extend a previous work regarding the application of machine learning techniques for design pattern detection, by adding a more extensive experimentation and enhancements in the analysis method. Here we exploit a combination of graph matching and machine learning techniques, implemented in a tool we developed, called MARPLE-DPD. Our approach allows the application of machine learning techniques, leveraging a modeling of design patterns that is able to represent pattern instances composed of a variable number of classes. We describe the experimentations for the detection of five design patterns on 10 open source software systems, compare the performances obtained by different learning models with respect to a baseline, and discuss the encountered issues.

Background: Dynamic aspects of gene regulatory networks are typically investigated by measuring system variables at multiple time points. Current state-of-the-art computational approaches for reconstructing gene networks directly build on such data, making a strong assumption that the system evolves in a synchronous fashion at fixed points in time. However, nowadays omics data are being generated with increasing time course granularity. Thus, modellers now have the possibility to represent the system as evolving in continuous time and to improve the models' expressiveness. Results: Continuous time Bayesian networks are proposed as a new approach for gene network reconstruction from time course expression data. Their performance was compared to two state-of-the-art methods: dynamic Bayesian networks and Granger causality analysis. On simulated data, the methods comparison was carried out for networks of increasing size, for measurements taken at different time granularity densities and for measurements unevenly spaced over time. Continuous time Bayesian networks outperformed the other methods in terms of the accuracy of regulatory interactions learnt from data for all network sizes. Furthermore, their performance degraded smoothly as the size of the network increased. Continuous time Bayesian networks were significantly better than dynamic Bayesian networks for all time granularities tested and better than Granger causality for dense time series. Both continuous time Bayesian networks and Granger causality performed robustly for unevenly spaced time series, with no significant loss of performance compared to the evenly spaced case, while the same did not hold true for dynamic Bayesian networks. The comparison included the IRMA experimental datasets which confirmed the effectiveness of the proposed method. Continuous time Bayesian networks were then applied to elucidate the regulatory mechanisms controlling murine T helper 17 (Th17) cell differentiation and were found to be effective in discovering well-known regulatory mechanisms, as well as new plausible biological insights. Conclusions: Continuous time Bayesian networks were effective on networks of both small and large size and were particularly feasible when the measurements were not evenly distributed over time. Reconstruction of the murine Th17 cell differentiation network using continuous time Bayesian networks revealed several autocrine loops, suggesting that Th17 cells may be auto regulating their own differentiation process.

In this paper the authors present a new option pricing scheme which deals with non constant volatility for the price of the underlying asset. The main feature, of the proposed pricing scheme, consists of exploiting recent developments, about Bayesian learning, within the Artificial Neural Networks framework. Indeed the Bayesian learning approach allows the data to speak for itself, i.e. to make a few general assumptions about the process to be modeled and to exploit all the available data, concerning the price of traded options, for modeling the implied volatility surface. The nonparametric model of the implied volatility surface, obtained through an Infinite Feedforward Neural Network and by exploiting the Bayesian formulation of the learning problem, is used whitin the proposed option pricing scheme. This pricing scheme relies upon the Dupire formula which maps the implied volatility surface to the corresponding local volatility function. Numerical experiments, for the case of the USD/DEM over-the-counter options, are presented together with a graphical analysis of the resulting smiles which witness the effectiveness of the overall approach to option pricing.

Streaming data are relevant to finance, computer science, and engineering while they are becoming increasingly important to medicine and biology. Continuous time Bayesian network classifiers are designed for analyzing multivariate streaming data when time duration of event matters. Structural and parametric learning for the class of continuous time Bayesian network classifiers are considered in the case where complete data is available. Conditional log-likelihood scoring is developed for structural learning on continuous time Bayesian network classifiers. Performance of continuous time Bayesian network classifiers learned when combining conditional log-likelihood scoring and Bayesian parameter estimation are compared with that achieved by continuous time Bayesian network classifiers when learning is based on marginal log-likelihood scoring and to that achieved by dynamic Bayesian network classifiers. Classifiers are compared in terms of accuracy and computation time. Comparison is based on numerical experiments where synthetic and real data are used. Results show that conditional log-likelihood scoring combined with Bayesian parameter estimation outperforms marginal log-likelihood scoring. Conditional log-likelihood scoring becomes even more effective when the amount of available data is limited. Continuous time Bayesian network classifiers outperform in terms of computation time and accuracy dynamic Bayesian network on synthetic and real data sets.

Continuous time Bayesian network classifiers are designed for analyzing multivariate streaming data when time duration of events matters. New continuous time Bayesian network classifiers are introduced while their conditional log-likelihood scoring function is developed. A learning algorithm, combining conditional log-likelihood with Bayesian parameter estimation is developed. Classification accuracies achieved on synthetic data by continuous time and dynamic Bayesian network classifiers are compared. Results show that conditional log-likelihood scor- ing combined with Bayesian parameter estimation outperforms marginal log-likelihood scoring in terms of classification accuracy. Continuous time Bayesian network classifiers are applied to post-stroke rehabilitation.

Dynamic aspects of regulatory networks are typically investigated by measuring relevant variables at multiple points in time. Current state-of-the-art approaches for gene network reconstruction directly build on such data, making the strong assumption that the system evolves in a synchronous fashion and in discrete time. However, omics data generated with increasing time-course granularity allow to model gene networks as systems whose state evolves in continuous time, thus improving the model’s expressiveness. In this work continuous time Bayesian networks are proposed as a new approach for regulatory network reconstruction from time-course expression data. Their performance is compared to that of two state-of-the-art methods: dynamic Bayesian networks and Granger causality. The comparison is accomplished using both simulated and experimental data. Continuous time Bayesian networks achieve the highest F-measure on both datasets. Furthermore, precision, recall and F-measure degrade in a smoother way than those of dynamic Bayesian networks and Granger causality, when the complexity of the gene regulatory network increases.

An approach that combines Self-Organizing maps, hierarchical clustering and network components is presented, aimed at comparing protein conformational ensembles obtained from multiple Molecular Dynamic simulations. As a first result the original ensembles can be summarized by using only the representative conformations of the clusters obtained. In addition the network components analysis allows to discover and interpret the dynamic behavior of the conformations won by each neuron. The results showed the ability of this approach to efficiently derive a functional interpretation of the protein dynamics described by the original conformational ensemble, highlighting its potential as a support for protein engineering., In Proceedings Wivace 2013, arXiv:1309.7122

Latent factor models have been proved to be the state of the art for the Collaborative Filtering approach in a Recommender System. However, latent factors obtained with mathematical methods applied to the user-item matrix can be hardly interpreted by humans. In this paper we exploit Topic Models applied to textual data associated with items to find explanations for latent factors. Based on the Movie Lens dataset and textual data about movies collected from Freebase we run a user study with over hundred participants to develop a reference dataset for evaluating different strategies towards more interpretable and portable latent factor models.

Portfolio analysis studies the impact of economic and financial scenarios on the performance of an investment portfolio, while portfolio optimization concerns asset allocation to achieve a trade-off between risk and return. In this chapter we exploit the interplay between modern portfolio theory and Bayesian networks to describe a new framework for portfolio analysis and optimization. Bayesian networks provide an effective way to interface models to data, allow efficient evidential reasoning, while their graphical language offers an intuitive interface by which the analyst can elicit his/her knowledge. The proposed framework leverages on evidential reasoning to understand the behavior of an investment portfolio in different economic and financial scenarios. It allows to formulate and solve a portfolio optimization problem, while coherently taking into account the investor’s market views. The Bayesian network framework for portfolio analysis and optimization is instantiated on the DJ Euro Stoxx 50 Index. Examples of portfolio analysis and optimization, exploiting evidential reasoning on Bayesian networks, are presented and discussed.