Steffen Pauws

A personalized music system for motivation in sport performance

We developed a personalized music system called IM4Sports (interactive music for sports) for individual exercising, although running is the prime target. Research prototype of the system consists of a personal computer, a portable music flash player, a heart sensor strap, and a pedometer.

Realization and user evaluation of an automatic playlist generator

An automatic music playlist generator called PATS (Personalized Automatic Track Selection) creates playlists that aim at suiting a particular listening situation. It uses dynamic clustering in which songs are grouped based on a weighted attribute-value similarity measure. An inductive learning algorithm is used to reveal the weights for attribute-values using user preference feedback. In a controlled user experiment, the quality of PATS-generated and randomly assembled playlists for jazz music was assessed in two listening situations. The two listening situations were “listening to soft music” and “listening to lively music.” Playlist quality was measured by precision (songs that suit the listening situation), coverage (songs that suit the listening situation but that were not already contained in previous playlists) and a rating score. Results showed that PATS playlists contained increasingly more preferred music (increasingly higher precision), covered more preferred music in the collection (higher coverage), and were rated higher than randomly assembled playlists.

Music playlist generation by adapted simulated annealing

We present the design of an algorithm for use in an interactive music system that automatically generates music playlists that fit the music preferences of a user. To this end, we introduce a formal model, define the problem of automatic playlist generation (APG), and prove its NP-hardness. We use a local search (LS) procedure employing a heuristic improvement to standard simulated annealing (SA) to solve the APG problem. In order to employ this LS procedure, we introduce an optimization variant of the APG problem, which includes the definition of penalty functions and a neighborhood structure. To improve upon the performance of the standard SA algorithm, we incorporated three heuristics referred to as song domain reduction, partial constraint voting, and a two-level neighborhood structure. We evaluate the developed algorithm by comparing it to a previously developed approach based on constraint satisfaction (CS), both in terms of run time performance and quality of the solutions. For the latter we not only considered the penalty of the resulting solutions, but we also performed a conclusive user evaluation to assess the subjective quality of the playlists generated by both algorithms. In all tests, the LS algorithm was shown to be a dramatic improvement over the CS algorithm.

Insightful stress detection from physiology modalities using Learning Vector Quantization

Abstract Stress in daily life can lead to severe conditions as burn-out and depression and has a major impact on society. Being able to measure mental stress reliably opens up the ability to intervene in an early stage. We performed a large-scale study in which skin conductance, respiration and electrocardiogram were measured in semi-controlled conditions. Using Learning Vector Quantization techniques, we obtained up to 88% accuracy in the classification task to separate stress from relaxation. Relevance learning was used to identify the most informative features, indicating that most information is embedded in the cardiac signals. In addition to commonly used features, we also explored various novel features, of which the very-high frequency band of the power spectrum was found to be a very relevant addition.

Robot Vacuum Cleaner Personality and Behavior

In this paper we report our study on the user experience of robot vacuum cleaner behavior. How do people want to experience this new type of cleaning appliance? Interviews were conducted to elicit a desired robot vacuum cleaner personality. With this knowledge in mind, behavior was designed for a future robot vacuum cleaner. A video prototype was used to evaluate how people experienced the behavior of this robot vacuum cleaner. The results indicate that people recognized the intended personality in the robot behavior. We recommend using a personality model as a tool for developing robot behavior.

Enriching music with synchronized lyrics, images and colored lights

We present a method to synchronize popular music with its lyrics at the stanza level. First we apply an algorithm to segment audio content into harmonically similar and/or contrasting progressions, i.e. the stanzas. We map the stanzas found to a sequence of labels, where stanzas with a similar progression are mapped to the same label. The lyrics are analyzed as well to compute a second sequence of labels. Using dynamic programming, an optimal match is found between the two sequences, resulting in a stanza-level synchronization of the lyrics and the audio. The synchronized lyrics can be used to compute a synchronized slide show to accompany the music, where the images are retrieved using the lyrics. For an additional enrichment of the experience, colored light effects are synchronized with the music that are computed from the sets of images. The song segmentation can be done reliably, while the mapping of the audio segments and lyrics gives encouraging results.

Proposals for enhanced health risk assessment and stratification in an integrated care scenario.

Objectives Population-based health risk assessment and stratification are considered highly relevant for large-scale implementation of integrated care by facilitating services design and case identification. The principal objective of the study was to analyse five health-risk assessment strategies and health indicators used in the five regions participating in the Advancing Care Coordination and Telehealth Deployment (ACT) programme (http://www.act-programme.eu). The second purpose was to elaborate on strategies toward enhanced health risk predictive modelling in the clinical scenario. Settings The five ACT regions: Scotland (UK), Basque Country (ES), Catalonia (ES), Lombardy (I) and Groningen (NL). Participants Responsible teams for regional data management in the five ACT regions. Primary and secondary outcome measures We characterised and compared risk assessment strategies among ACT regions by analysing operational health risk predictive modelling tools for population-based stratification, as well as available health indicators at regional level. The analysis of the risk assessment tool deployed in Catalonia in 2015 (GMAs, Adjusted Morbidity Groups) was used as a basis to propose how population-based analytics could contribute to clinical risk prediction. Results There was consensus on the need for a population health approach to generate health risk predictive modelling. However, this strategy was fully in place only in two ACT regions: Basque Country and Catalonia. We found marked differences among regions in health risk predictive modelling tools and health indicators, and identified key factors constraining their comparability. The research proposes means to overcome current limitations and the use of population-based health risk prediction for enhanced clinical risk assessment. Conclusions The results indicate the need for further efforts to improve both comparability and flexibility of current population-based health risk predictive modelling approaches. Applicability and impact of the proposals for enhanced clinical risk assessment require prospective evaluation.

Heuristic Approaches for the Quartet Method of Hierarchical Clustering

Given a set of objects and their pairwise distances, we wish to determine a visual representation of the data. We use the quartet paradigm to compute a hierarchy of clusters of the objects. The method is based on an NP-hard graph optimization problem called the Minimum Quartet Tree Cost problem. This paper presents and compares several heuristic approaches to approximate the optimal hierarchy. The performance of the algorithms is tested through extensive computational experiments and it is shown that the Reduced Variable Neighborhood Search heuristic is the most effective approach to the problem, obtaining high-quality solutions in short computational running times.

Depression as an independent prognostic factor for all-cause mortality after a hospital admission for worsening heart failure

BACKGROUND Depression is associated with increased mortality amongst patients with chronic heart failure (HF). Whether depression is an independent predictor of outcome in patients admitted for worsening of HF is unclear. METHODS OPERA-HF is an observational study enrolling patients hospitalized with worsening HF. Depression was assessed by the Hospital Anxiety and Depression Scale (HADS-D) questionnaire. Comorbidity was assessed by the Charlson Comorbidity Index (CCI). Kaplan-Meier and Cox regression analyses were used to estimate the association between depression and all-cause mortality. RESULTS Of 242 patients who completed the HADS-D questionnaire, 153, 54 and 35 patients had no (score 0-7), mild (score 8-10) or moderate-to-severe (score 11-21) depression, respectively. During follow-up, 35 patients died, with a median time follow-up of 360days amongst survivors (interquartile range, IQR 217-574days). In univariable analysis, moderate-to-severe depression was associated with an increased risk of death (HR: 4.9; 95% CI: 2.3 to 10.2; P<0.001) compared to no depression. Moderate-to-severe depression also predicted all-cause mortality after controlling for age, CCI score, NYHA class IV, NT-proBNP and treatment with mineralocorticoid receptor antagonist, beta-blocker and diuretics (HR: 3.0; 95% CI: 1.3 to 7.0; P<0.05). CONCLUSIONS Depression is strongly associated with an adverse outcome in the year following discharge after an admission to hospital for worsening HF. The association is only partly explained by the severity of HF or comorbidity. Further research is required to demonstrate whether recognition and treatment of depression improves patient outcomes.

Towards Emotion Classification Using Appraisal Modeling

The authors studied whether a two-step approach based on appraisal modeling could help in improving performance of emotion classification from sensor data that is typically executed in a one-stage approach in which sensor data is directly classified into a discrete emotion label. The proposed intermediate step is inspired by appraisal models in which emotions are characterized using appraisal dimensions, and subdivides the task in a person-dependent and person-independent stage. In this paper, the authors assessed feasibility of this second stage: the classification of emotion from appraisal data. They applied a variety of machine learning techniques and used visualization techniques to gain further insight into the classification task. Appraisal theory assumes the second step to be independent of the individual. Results obtained are promising, but do indicate that not all emotions can be equally well classified, perhaps indicating that the second stage is not as person-independent as proposed in the literature.