In this article we propose a conceptual framework to study ensembles of conformal predictors (CP), that we call Ensemble Predictors (EP). Our approach is inspired by the application of imprecise probabilities in information fusion. Based on the proposed framework, we study, for the first time in the literature, the theoretical properties of CP ensembles in a general setting, by focusing on simple and commonly used possibilistic combination rules. We also illustrate the applicability of the proposed methods in the setting of multivariate time-series classification, showing that these methods provide better performance (in terms of both robustness, conservativeness, accuracy and running time) than both standard classification algorithms and other combination rules proposed in the literature, on a large set of benchmarks from the UCR time series archive.

The recognition and evaluation of emotional states have important applications in the medical domain. Emotion recognition has been an active area of research in recent years, with a significant focus on data sources such as images or text, but also physiological signals like the electrocardiogram or electroencephalogram. However, traditional data collection and labelling methods often rely on exaggeratedly acted emotions or intense real emotions triggered by strong stimuli. These fail to mimic the conditions of most real world scenarios, where highly-intense emotional manifestations are not the most common. This work addresses these issues by evaluating the suitability of using physiological signals for emotion classification. We conducted a data collection protocol to acquire realistic emotional data in an isolated and undisturbed setup, with exposure to weak stimuli and standard activities, to capture low-intensity emotional manifestations. We present an initial exploratory analysis of physiological data, followed by the development of an emotion recognition training strategy using Machine Learning algorithms. Our approach achieved an optimal balanced accuracy of 47.97% and an area under the receiver operating characteristic curve of 72.09% for a multi-class classification problem with four emotion classes. Although the results may seem modest at a first glance, it is important to consider the inherent difficulty of distinguishing between subtle, low-intensity emotions, as well as the relevance of the problem to healthcare applications. Therefore, this work supports the development of more holistic, patient-centred healthcare solutions with emotion recognition.

Work-related disorders are a growing issue for office workers and represent a significant burden to public health. Work aspects such as sitting for prolonged periods and occupational stress are modifiable risk factors highly associated with occupational disorders in office workers. The PrevOccu-pAI Project (Prevention of Occupational Disorders in Public Administrations based on Artificial Intelligence) objectively investigates relationships between a variety of occupational risk factors and physiological outcomes. For this purpose, a data acquisition protocol was carried out at the Portuguese Tax and Customs Authority. Physiological, movement, and environmental signals from office workers were acquired during five consecutive workdays using a smartphone, a smartwatch, and two electromyography sensors. Additionally, demographic, occupational, and pain information were collected through questionnaires. The present manuscript provides a detailed description of the PrevOccupAI acquisition protocol. The collected data is used to gather knowledge regarding modifiable factors at the individual and organisational levels.

In this paper, we study human-AI collaboration protocols, a design-oriented construct aimed at establishing and evaluating how humans and AI can collaborate in cognitive tasks. We applied this construct in two user studies involving 12 specialist radiologists (the knee MRI study) and 44 ECG readers of varying expertise (the ECG study), who evaluated 240 and 20 cases, respectively, in different collaboration configurations. We confirm the utility of AI support but find that XAI can be associated with a "white-box paradox", producing a null or detrimental effect. We also find that the order of presentation matters: AI-first protocols are associated with higher diagnostic accuracy than human-first protocols, and with higher accuracy than both humans and AI alone. Our findings identify the best conditions for AI to augment human diagnostic skills, rather than trigger dysfunctional responses and cognitive biases that can undermine decision effectiveness.

Time series similarity matrices (informally, recurrence plots), are useful tools for time series data mining. They can be used to guide data exploration, and various useful features can be derived from them and then fed into downstream analytics. However, time series similarity matrices suffer from very poor scalability, taxing both time and memory requirements. In this work, we introduce novel ideas that allow us to scale the largest time series similarity matrices that can be examined by several orders of magnitude. The first idea is a novel algorithm to compute the matrices in a way that removes dependency on the subsequence length. This algorithm is so fast that it allows us to now address datasets where the memory limitations begin to dominate. Our second novel contribution is a multiscale algorithm that computes an approximation of the matrix appropriate for the limitations of the user’s memory/screen-resolution, then performs a local, just-in-time recomputation of any region that the user wishes to zoom-in on. Given that we can largely remove time and space barriers, human visual attention then becomes the bottleneck. We further introduce algorithms that search massive matrices with quadrillions of cells and then prioritize regions for later attention by either humans or algorithms. We will demonstrate the utility of our ideas for data exploration, segmentation, and classification in diverse domains.

In recent times, infrastructure-free indoor positioning has been an important topic of research. Many of the proposed systems are based on pedestrian dead reckoning, thus relying on estimating the heading of the pedestrian. While many studies successfully address the problem of estimating the heading of the device, current approaches have the limitation of requiring the device to be aligned with the pedestrian. To address this problem, we propose an algorithm for estimating the misalignment between the pedestrian and the device by evaluating the acceleration data fit a simplified gait model in each direction. Contrary to similar algorithms, the proposal in this paper does not require a previously trained model nor the detection of steps, and can be implemented using only acceleration data. Furthermore, our experimental results show a significant improvement over the current state of the art.

Nowadays, data scientists are capable of manipulating and extracting complex information from time series data, given the current diversity of tools at their disposal. However, the plethora of tools that target data exploration and pattern search may require an extensive amount of time to develop methods that correspond to the data scientist's reasoning, in order to solve their queries. The development of new methods, tightly related with the reasoning and visual analysis of time series data, is of great relevance to improving complexity and productivity of pattern and query search tasks. In this work, we propose a novel tool, capable of exploring time series data for pattern and query search tasks in a set of 3 symbolic steps: Pre-Processing, Symbolic Connotation and Search. The framework is called SSTS (Symbolic Search in Time Series) and uses regular expression queries to search the desired patterns in a symbolic representation of the signal. By adopting a set of symbolic methods, this approach has the purpose of increasing the expressiveness in solving standard pattern and query tasks, enabling the creation of queries more closely related to the reasoning and visual analysis of the signal. We demonstrate the tool's effectiveness by presenting 9 examples with several types of queries on time series. The SSTS queries were compared with standard code developed in Python, in terms of cognitive effort, vocabulary required, code length, volume, interpretation and difficulty metrics based on the Halstead complexity measures. The results demonstrate that this methodology is a valid approach and delivers a new abstraction layer on data analysis of time series.

When a comparison between time series is required, measurement functions provide meaningful scores to characterize similarity between sequences. Quite often, time series appear warped in time, i.e, although they may exhibit amplitude and shape similarity, they appear dephased in time. The most common algorithm to overcome this challenge is the Dynamic Time Warping, which aligns each sequence prior establishing distance measurements. However, Dynamic Time Warping takes only into account amplitude similarity. A distance which characterizes the degree of time warping between two sequences can deliver new insights for applications where the timing factor is essential, such well-defined movements during sports or rehabilitation exercises. We propose a novel measurement called Time Alignment Measurement, which delivers similarity information on the temporal domain. We demonstrate the potential of our approach in measuring performance of time series alignment methodologies and in the characterization of synthetic and real time series data acquired during human movement.

As sensor-rich mobile devices became a commodity, more opportunities appeared for the creation of location-aware services. While GPS is a well established solution for outdoor localization, there is still no standard solution for localization indoors. This paper presents a novel accurate indoor positioning mechanism that is meant to run in common smartphones to be a readily and widely available solution. The system is based on multiple gait-model based filtering techniques for accurate movement quantification in combination with an advanced fused positioning mechanism that leverages sequences of opportunistic observations towards an accurate localization process. Magnetic field fluctuations, Wi-Fi readings and movement data are incrementally matched with a feature spot map containing multi-dimensional spatially-related features that characterize the building. A novel and convenient way of mapping the architectural and environmental properties of buildings is also introduced, which avoids the burden normally associated with the process. The system has been evaluated by multiple users in open and crowded spaces where overall median localization errors between 1.11 m and 1.68 m were obtained. While the reported errors are already satisfactory in the context of indoor localization, improvements may be readily achieved through the inclusion of additional reference features. High accuracy performance coupled with an opportunistic and infrastructure-free approach creates a very desirable solution for the indoor localization market doge.

The prospect of connecting the brain and body to a technological device can elicit a broad range of responses from potential users. Early adopters are thrilled by the possibility of a device that can interface directly to the human nervous system. For the vast majority, interest is tempered by caution, as nascent varieties of physiological computing systems raise as many questions as answers about how we will interact with computers in the future.

Electromyographic (EMG) signals play a key role in many clinical and biomedical applications. They can be used for identifying patients with muscular disabilities, assessing lower-back pain, kinesiology and motor control. There are three common applications of the EMG signal: (1) to determine the activation timing of the muscle; (2) to estimate the force produced by the muscle and (3) to analyze muscular fatigue through analysis of the frequency spectrum of the signal. We have developed an EMG tool that was incorporated in an existing web-based biosignal acquisition and processing framework. This tool can be used on a post-processing environment and provides not only frequency and time parameters, but also an automatic detection of starting and ending times for muscular voluntary contractions using a threshold-based algorithm with the inclusion of the Teager–Kaiser energy operator. The algorithm for the muscular voluntary contraction detection can also be reported after a real-time acquisition, in order to discard possible outliers and simultaneously compare activation times in different muscles. This tool covers all known applications and allows a careful and detailed analysis of the EMG signal for both clinicians and researchers. The detection algorithm works without user interference and is also user-independent. It manages to detect muscular activations in an interactive process. The user simply has to select the signal's time interval as input, and the outcomes are provided afterwards. RESEARCH HIGHLIGHTS An interactive analysis tool for electromyographic (EMG) signals has been developed. The tool gives the user control over signal processing algorithms, enabling human–computer interaction and provides visual information from the signals and processing results. This tool differs from the existing ones due to the inclusion of an automatic detection algorithm for the muscular voluntary contraction—threshold-based method with the inclusion of the Teager–Keaser energy operator. It allows a careful and detailed analysis of EMG signals both for clinicians and researchers.

We address the problem of classification of data with low separability. We adopt a Bayesian approach, with discriminant functions expressing a posteriori class probabilities. We propose a novel classification scheme incorporating classification error probability estimates in the decision process. We extend this approach into a classifier fusion framework. Presented methods are evaluated in the context of user authentication, using multimodal biometrics. Results on real data confirm the usefulness of the proposed method, outperforming the corresponding deterministic classifier.