Search

Cross-lingual text representations have gained popularity lately and act as the backbone of many tasks such as unsupervised machine translation and cross-lingual information retrieval, to name a few. However, evaluation of such representations is difficult in the domains beyond standard benchmarks due to the necessity of obtaining domain-specific parallel language data across different pairs of languages. In this paper, we propose an automatic metric for evaluating the quality of cross-lingual textual representations using images as a proxy in a paired image-text evaluation dataset. Experimentally, Backretrieval is shown to highly correlate with ground truth metrics on annotated datasets, and our analysis shows statistically significant improvements over baselines. Our experiments conclude with a case study on a recipe dataset without parallel cross-lingual data. We illustrate how to judge cross-lingual embedding quality with Backretrieval, and validate the outcome with a small human study., Comment: SIGIR 2021

One of the main shortcomings of received signal strength-based indoor localisation techniques is the labour and time cost involved in acquiring labelled `ground-truth' training data. This training data is often obtained through fingerprinting, which involves visiting all prescribed locations to capture sensor observations throughout the environment. In this work, the authors present a helmet for localisation optimisation (H4LO): a low-cost robotic system designed to cut down on said labour by utilising an off-the-shelf light detection and ranging device. This system allows for simultaneous localisation and mapping, providing the human user with accurate pose estimation and a corresponding map of the environment. The high-resolution location estimation can then be used to train a positioning model, where received signal strength data is acquired from a human-worn wearable device. The method is evaluated using live measurements, recorded within a residential property. They compare the groundtruth location labels generated automatically by the H4LO system with a camera-based fingerprinting technique from previous work. They find that the system remains comparable in performance to the less efficient camera-based method, whilst removing the need for time-consuming labour associated with registering the user's location.

Active Learning (AL) refers to the setting where the learner has the ability to perform queries to an oracle to acquire the true label of an instance or, sometimes, a set of instances. Even though Active Learning has been studied extensively, the setting is usually restricted to assume that the oracle is trustworthy and will provide the actual label. We argue that, while common, this approach can be made more flexible to account for different forms of supervision. In this paper, we propose a new framework that allows the algorithm to request the label for a bag of samples at a time. Although this label will come in the form of proportions of class labels in the bags and therefore encode less information, we demonstrate that we can still learn effectively.

Delivering effortless interactions and appropriate interventions through pervasive systems requires making sense of multiple streams of sensor data. This is particularly challenging when these concern people&rsquo, s natural behaviours in the real world. This paper takes a multidisciplinary perspective of annotation and draws on an exploratory study of 12 people, who were encouraged to use a multi-modal annotation app while living in a prototype smart home. Analysis of the app usage data and of semi-structured interviews with the participants revealed strengths and limitations regarding self-annotation in a naturalistic context. Handing control of the annotation process to research participants enabled them to reason about their own data, while generating accounts that were appropriate and acceptable to them. Self-annotation provided participants an opportunity to reflect on themselves and their routines, but it was also a means to express themselves freely and sometimes even a backchannel to communicate playfully with the researchers. However, self-annotation may not be an effective way to capture accurate start and finish times for activities, or location associated with activity information. This paper offers new insights and recommendations for the design of self-annotation tools for deployment in the real world.

The SPHERE project is devoted to advancing eHealth in a smart-home context, and supports full-scale sensing and data analysis to enable a generic healthcare service. We describe, from a data-science perspective, our experience of taking the system out of the laboratory into more than thirty homes in Bristol, UK. We describe the infrastructure and processes that had to be developed along the way, describe how we train and deploy Machine Learning systems in this context, and give a realistic appraisal of the state of the deployed systems.

There's a widely known need to revise current forms of healthcare provision. Of particular interest are sensing systems in the home, which have been central to several studies. This article presents an overview of this rapidly growing body of work, as well as the implications for machine learning, with an aim of uncovering the gap between the state of the art and the broad needs of healthcare services in ambient assisted living. Most approaches address specific healthcare concerns, which typically result in solutions that aren't able to support full-scale sensing and data analysis for a more generic healthcare service, but the approach in this article differs from seamlessly linking multimodel data-collecting infrastructure and data analytics together in an AAL platform. This article also outlines a multimodality sensor platform with heterogeneous network connectivity, which is under development in the sensor platform for healthcare in a residential environment (SPHERE) Interdisciplinary Research Collaboration (IRC).

There has been significant recent interest in sensing systems and ‘smart environments’, with a number of longitudinal studies in this area. Typically the goal of these studies is to develop methods to predict, at any one moment of time, the activity or activities that the resident(s) of the home are engaged in, which may in turn be used for determining normal or abnormal patterns of behaviour (e.g. in a health-care setting). Classification algorithms, such as Conditional Random Field (CRFs), typically consider sensor activations as features but these are often treated as if they were independent, which in general they are not. Our hypothesis is that learning patterns based on combinations of sensors will be more powerful than single sensors alone. The exhaustive approach – to take all possible combinations of sensors and learn classifier weights for each combination – is clearly computationally prohibitive. We show that through the application of signal processing and information-theoretic techniques we can learn about the sensor topology in the home (i.e. learn an adjacency matrix) which enables us to determine the combinations of sensors that will be useful for classification ahead of time. As a result we can achieve classification performance better than that of the exhaustive approach, whilst only incurring a small cost in terms of computational resources. We demonstrate our results on several datasets, showing that our method is robust in terms of variations in the layout and the number of residents in the house. Furthermore, we have incorporated the adjacency matrix into the CRF learning framework and have shown that it can improve performance over multiple baselines.

It can be tempting to think about smart homes like one thinks about smart cities. On the surface, smart homes and smart cities comprise coherent systems enabled by similar sensing and interactive technologies. It can also be argued that both are broadly underpinned by shared goals of sustainable development, inclusive user engagement and improved service delivery. However, the home possesses unique characteristics that must be considered in order to develop effective smart home systems that are adopted in the real world [37].

We introduce and analyse a flexible and efficient implementation of Bayesian dictionary learning for sparse coding. By placing Gaussian-inverse-Gamma hierarchical priors on the coefficients, the model can automatically determine the required sparsity level for good reconstructions, whilst also automatically learning the noise level in the data, obviating the need for heuristic methods for choosing sparsity levels. This model can be solved efficiently using Variational Message Passing (VMP), which we have implemented in the Infer.NET framework for probabilistic programming and inference. We analyse the properties of the model via empirical validation on several accelerometer datasets. We provide source code to replicate all of the experiments in this paper.

In residential environments, floor maps, often required by location-based services, cannot be trivially acquired. Researchers have addressed the problem of automatic floor map construction in indoor environments using various modalities, such as inertial sensors, Radio Frequency (RF) fingerprinting and video cameras. Considering that some of these techniques are unavailable or impractical to implement in residential environments, in this paper, we focus on using RF signals to predict the number of walls between a wearable device and an access point. Using both supervised and unsupervised learning techniques on two data sets; a system-level data set of Bluetooth packets, and measurements on the signal attenuation, we construct wall prediction models that yield up to 91% identification rate. As a proof-of-concept, we also use the wall prediction models to infer the floor plan of a smart home deployment in a real residential environment.

This paper presents an automated framework for the detection of the QRS complex from Electrocardiogram (ECG) signals. We introduce an artefact-tolerant pre-processing algorithm which emphasises a number of characteristics of the ECG that are representative of the QRS complex. With this processed ECG signal we train Logistic Regression and Support Vector Machine classification models. With our approach we obtain over 99.7% detection sensitivity and precision on the MIT-BIH database without using supplementary de-noising or pre-emphasis filters.

This paper investigates the fully automated computer-based detection of allergic reaction in oral food challenges using pediatric ECG signals. Nonallergic background is modeled using a mixture of Gaussians during oral food challenges, and the model likelihoods are used to determine whether a subject is allergic to a food type. The system performance is assessed on the dataset of 24 children (15 allergic and 9 nonallergic) totaling 34 h of data. The proposed detector correctly classified all nonallergic subjects (100 % specificity) and 12 allergic subjects (80 % sensitivity) and is capable of detecting allergy on average 17 min earlier than trained clinicians during oral food challenges, the gold standard of allergy diagnosis. Inclusion of the developed allergy classification platform during oral food challenges recorded would result in a 30% reduction of doses administered to allergic subjects. The results of study introduce the possibility to halt challenges earlier which can safely advance the state of clinical art of allergy diagnosis by reducing the overall exposure to the allergens.

In this paper, we tackle the problem of the food allergic detection in children, based on the analysis of the ECG signal. Through the detection of some changes of this signal, it is possible to detect any reaction before the tested subject experiments any physical reaction or any reaction that could be harmful to his/her life. To be able to realize this process in real-time and with portable devices, it is necessary to reduce the computational cost of the full process, from the ECG analysis to the allergy detection process.

This paper investigates the automated classification of oral food challenges ('allergy tests'). The electrocardiograms (ECG) of the subjects being tested for allergies were recorded via a wireless mote, and the QRS complexes were manually annotated and 18 features were extracted from the signals. Principal component analysis was used for feature decorelation and dimensionality reduction and diagonal covariance Gaussian mixture models were used to model non-reaction baseline patient condition. The generated subject independent log likelihood plots were used to separate allergic reaction by means of subject adaptive thresholding. The platform resulted in 87% accuracy of classification with 100% specificity. The algorithm presented can detect allergy up to 30 minutes sooner than the current state of the clinical art allergy detection (7minutes ± 9).

This paper details the process of oral food challenges (‘allergy tests’) and steps followed to investigate whether automatic classification of the tests is possible. It has been observed by trained staff that the mood and physiological signals of a subject being tested for allergies can change during the test if they are sensitive to the allergen they are being tested against. Data from thirteen subjects was recorded, and thirteen features were extracted from each of these datasets. The changes in the features were then analysed over the course of each test. It was noted that when a subject failed the challenge, some of the features extracted from the ECG trace changed suddenly near the time that the test was stopped. Threshold classification was employed, and ROC curves were generated. Some features gave rise to ROC areas of over 0.97 on certain subjects. An average ROC area of 0.57 was computed over all subjects and all features due to wide inter subject variability.

There is no uniform approach in the literature for modelling sequential correlations in sequence classification problems. It is easy to find examples of unstructured models (e.g. logistic regression) where correlations are not taken into account at all, but there are also many examples where the correlations are explicitly incorporated into a – potentially computationally expensive – structured classification model (e.g. conditional random fields). In this paper we lay theoretical and empirical foundations for clarifying the types of problem which necessitate direct modelling of correlations in sequences, and the types of problem where unstructured models that capture sequential aspects solely through features are sufficient. The theoretical work in this paper shows that the rate of decay of auto-correlations within a sequence is related to the excess classification risk that is incurred by ignoring the structural aspect of the data. This is an intuitively appealing result, demonstrating the intimate link between the auto-correlations and excess classification risk. Drawing directly on this theory, we develop well-founded visual analytics tools that can be applied a priori on data sequences and we demonstrate how these tools can guide practitioners in specifying feature representations based on auto-correlation profiles. Empirical analysis is performed on three sequential datasets. With baseline feature templates, structured and unstructured models achieve similar performance, indicating no initial preference for either model. We then apply the visual analytics tools to the datasets, and show that classification performance in all cases is improved over baseline results when our tools are involved in defining feature representations.

This paper describes the performance of beat detection and heart rate variability (HRV) feature extraction on electrocardiogram signals which have been compressed and reconstructed with a lossy compression algorithm. The set partitioning in hierarchical trees (SPIHT) compression algorithm was used with sixteen compression ratios (CR) between 2 and 50 over the records of the MIT/BIH arrhythmia database. Sensitivities and specificities between 99% and 85% were computed for each CR utilised. The extracted HRV features were between 99% and 82% similar to the features extracted from the annotated records. A notable accuracy drop over all features extracted was noted beyond a CR of 30, with falls of 10% accuracy beyond this compression ratio.

This paper serves as a survey and empirical evaluation of the state-of-the-art in activity recognition methods using accelerometers. The paper is particularly focused on long-term activity recognition in real-world settings. In these environments, data collection is not a trivial matter, thus, there are performance trade-offs between prediction accuracy, which is not the sole system objective, and keeping the maintenance overhead at minimum levels. We examine research that has focused on the selection of activities, the features that are extracted from the accelerometer data, the segmentation of the time-series data, the locations of accelerometers, the selection and configuration trade-offs, the test/retest reliability, and the generalisation performance. Furthermore, we study these questions from an experimental platform and show, somewhat surprisingly, that many disparate experimental configurations yield comparable predictive performance on testing data. Our understanding of these results is that the experimental setup directly and indirectly defines a pathway for context to be delivered to the classifier, and that, in some settings, certain configurations are more optimal than alternatives. We conclude by identifying how the main results of this work can be used in practice, specifically in experimental configurations in challenging experimental conditions.