A Hierarchical Multitask Learning Approach for the Recognition of Activities of Daily Living Using Data from Wearable Sensors.

Machine learning with deep neural networks (DNNs) is widely used for human activity recognition (HAR) to automatically learn features, identify and analyze activities, and to produce a consequential outcome in numerous applications. However, learning robust features requires an enormous number of labeled data. Therefore, implementing a DNN either requires creating a large dataset or needs to use the pre-trained models on different datasets. Multitask learning (MTL) is a machine learning paradigm where a model is trained to perform multiple tasks simultaneously, with the idea that sharing information between tasks can lead to improved performance on each individual task. This paper presents a novel MTL approach that employs combined training for human activities with different temporal scales of atomic and composite activities. Atomic activities are basic, indivisible actions that are readily identifiable and classifiable. Composite activities are complex actions that comprise a sequence or combination of atomic activities. The proposed MTL approach can help in addressing challenges related to recognizing and predicting both atomic and composite activities. It can also help in providing a solution to the data scarcity problem by simultaneously learning multiple related tasks so that knowledge from each task can be reused by the others. The proposed approach offers advantages like improved data efficiency, reduced overfitting due to shared representations, and fast learning through the use of auxiliary information. The proposed approach exploits the similarities and differences between multiple tasks so that these tasks can share the parameter structure, which improves model performance. The paper also figures out which tasks should be learned together and which tasks should be learned separately. If the tasks are properly selected, the shared structure of each task can help it learn more from other tasks.

Sleep Stage Classification in Children Using Self-Attention and Gaussian Noise Data Augmentation.

The analysis of sleep stages for children plays an important role in early diagnosis and treatment. This paper introduces our sleep stage classification method addressing the following two challenges: the first is the data imbalance problem, i.e., the highly skewed class distribution with underrepresented minority classes. For this, a Gaussian Noise Data Augmentation (GNDA) algorithm was applied to polysomnography recordings to seek the balance of data sizes for different sleep stages. The second challenge is the difficulty in identifying a minority class of sleep stages, given their short sleep duration and similarities to other stages in terms of EEG characteristics. To overcome this, we developed a DeConvolution- and Self-Attention-based Model (DCSAM) which can inverse the feature map of a hidden layer to the input space to extract local features and extract the correlations between all possible pairs of features to distinguish sleep stages. The results on our dataset show that DCSAM based on GNDA obtains an accuracy of 90.26% and a macro F1-score of 86.51% which are higher than those of our previous method. We also tested DCSAM on a well-known public dataset-Sleep-EDFX-to prove whether it is applicable to sleep data from adults. It achieves a comparable performance to state-of-the-art methods, especially accuracies of 91.77%, 92.54%, 94.73%, and 95.30% for six-stage, five-stage, four-stage, and three-stage classification, respectively. These results imply that our DCSAM based on GNDA has a great potential to offer performance improvements in various medical domains by considering the data imbalance problems and correlations among features in time series data.

Explainable Artificial Intelligence (XAI) in Pain Research: Understanding the Role of Electrodermal Activity for Automated Pain Recognition.

Artificial intelligence and especially deep learning methods have achieved outstanding results for various applications in the past few years. Pain recognition is one of them, as various models have been proposed to replace the previous gold standard with an automated and objective assessment. While the accuracy of such models could be increased incrementally, the understandability and transparency of these systems have not been the main focus of the research community thus far. Thus, in this work, several outcomes and insights of explainable artificial intelligence applied to the electrodermal activity sensor data of the PainMonit and BioVid Heat Pain Database are presented. For this purpose, the importance of hand-crafted features is evaluated using recursive feature elimination based on impurity scores in Random Forest (RF) models. Additionally, Gradient-weighted class activation mapping is applied to highlight the most impactful features learned by deep learning models. Our studies highlight the following insights: (1) Very simple hand-crafted features can yield comparative performances to deep learning models for pain recognition, especially when properly selected with recursive feature elimination. Thus, the use of complex neural networks should be questioned in pain recognition, especially considering their computational costs; and (2) both traditional feature engineering and deep feature learning approaches rely on simple characteristics of the input time-series data to make their decision in the context of automated pain recognition.

A new pairwise deep learning feature for environmental microorganism image analysis.

Environmental microorganism (EM) offers a highly efficient, harmless, and low-cost solution to environmental pollution. They are used in sanitation, monitoring, and decomposition of environmental pollutants. However, this depends on the proper identification of suitable microorganisms. In order to fasten, lower the cost, and increase consistency and accuracy of identification, we propose the novel pairwise deep learning features (PDLFs) to analyze microorganisms. The PDLFs technique combines the capability of handcrafted and deep learning features. In this technique, we leverage the Shi and Tomasi interest points by extracting deep learning features from patches which are centered at interest points' locations. Then, to increase the number of potential features that have intermediate spatial characteristics between nearby interest points, we use Delaunay triangulation theorem and straight line geometric theorem to pair the nearby deep learning features. The potential of pairwise features is justified on the classification of EMs using SVMs, Linear discriminant analysis, Logistic regression, XGBoost and Random Forest classifier. The pairwise features obtain outstanding results of 99.17%, 91.34%, 91.32%, 91.48%, and 99.56%, which are the increase of about 5.95%, 62.40%, 62.37%, 61.84%, and 3.23% in accuracy, F1-score, recall, precision, and specificity respectively, compared to non-paired deep learning features.

Sleep stage classification for child patients using DeConvolutional Neural Network.

Studies from the literature show that the prevalence of sleep disorder in children is far higher than that in adults. Although much research effort has been made on sleep stage classification for adults, children have significantly different characteristics of sleep stages. Therefore, there is an urgent need for sleep stage classification targeting children in particular. Our method focuses on two issues: The first is timestamp-based segmentation (TSS) to deal with the fine-grained annotation of sleep stage labels for each timestamp. Compared to this, popular sliding window approaches unnecessarily aggregate such labels into coarse-grained ones. We utilize DeConvolutional Neural Network (DCNN) that inversely maps features of a hidden layer back to the input space to predict the sleep stage label at each timestamp. Thus, our DCNN can yield better classification performances by considering labels at numerous timestamps. The second issue is the necessity of multiple channels. Different clinical signs, symptoms or other auxiliary examinations could be represented by different Polysomnography (PSG) recordings, so all of them should be analyzed comprehensively. We therefor exploit multivariate time-series of PSG recordings, including 6 electroencephalograms (EEGs) channels, 2 electrooculograms (EOGs) channels (left and right), 1 electromyogram (chin EMG) channel and two leg electromyogram channels. Our DCNN-based method is tested on our SDCP dataset collected from child patients aged from 5 to 10 years old. The results show that our method yields the overall classification accuracy of 84.27% and macro F1-score of 72.51% which are higher than those of existing sliding window-based methods. One of the biggest advantages of our DCNN-based method is that it processes raw PSG recordings and internally extracts features useful for accurate sleep stage classification. We examine whether this is applicable for sleep data of adult patients by testing our method on a well-known public dataset Sleep-EDFX. Our method achieves the average overall accuracy of 90.89% which is comparable to those of state-of-the-art methods without using any hand-crafted features. This result indicates the great potential of our method because it can be generally used for timestamp-level classification on multivariate time-series in various medical fields. Additionally, we provide source codes so that researchers can reproduce the results in this paper and extend our method.

Deep Transfer Learning for Time Series Data Based on Sensor Modality Classification.

The scarcity of labelled time-series data can hinder a proper training of deep learning models. This is especially relevant for the growing field of ubiquitous computing, where data coming from wearable devices have to be analysed using pattern recognition techniques to provide meaningful applications. To address this problem, we propose a transfer learning method based on attributing sensor modality labels to a large amount of time-series data collected from various application fields. Using these data, our method firstly trains a Deep Neural Network (DNN) that can learn general characteristics of time-series data, then transfers it to another DNN designed to solve a specific target problem. In addition, we propose a general architecture that can adapt the transferred DNN regardless of the sensors used in the target field making our approach in particular suitable for multichannel data. We test our method for two ubiquitous computing problems-Human Activity Recognition (HAR) and Emotion Recognition (ER)-and compare it a baseline training the DNN without using transfer learning. For HAR, we also introduce a new dataset, Cognitive Village-MSBand (CogAge), which contains data for 61 atomic activities acquired from three wearable devices (smartphone, smartwatch, and smartglasses). Our results show that our transfer learning approach outperforms the baseline for both HAR and ER.

Rank Pooling Approach for Wearable Sensor-Based ADLs Recognition.

This paper addresses wearable-based recognition of Activities of Daily Living (ADLs) which are composed of several repetitive and concurrent short movements having temporal dependencies. It is improbable to directly use sensor data to recognize these long-term composite activities because two examples (data sequences) of the same ADL result in largely diverse sensory data. However, they may be similar in terms of more semantic and meaningful short-term atomic actions. Therefore, we propose a two-level hierarchical model for recognition of ADLs. Firstly, atomic activities are detected and their probabilistic scores are generated at the lower level. Secondly, we deal with the temporal transitions of atomic activities using a temporal pooling method, rank pooling. This enables us to encode the ordering of probabilistic scores for atomic activities at the higher level of our model. Rank pooling leads to a 5-13% improvement in results as compared to the other popularly used techniques. We also produce a large dataset of 61 atomic and 7 composite activities for our experiments.

Comparison of Feature Learning Methods for Human Activity Recognition Using Wearable Sensors.

Getting a good feature representation of data is paramount for Human Activity Recognition (HAR) using wearable sensors. An increasing number of feature learning approaches-in particular deep-learning based-have been proposed to extract an effective feature representation by analyzing large amounts of data. However, getting an objective interpretation of their performances faces two problems: the lack of a baseline evaluation setup, which makes a strict comparison between them impossible, and the insufficiency of implementation details, which can hinder their use. In this paper, we attempt to address both issues: we firstly propose an evaluation framework allowing a rigorous comparison of features extracted by different methods, and use it to carry out extensive experiments with state-of-the-art feature learning approaches. We then provide all the codes and implementation details to make both the reproduction of the results reported in this paper and the re-use of our framework easier for other researchers. Our studies carried out on the OPPORTUNITY and UniMiB-SHAR datasets highlight the effectiveness of hybrid deep-learning architectures involving convolutional and Long-Short-Term-Memory (LSTM) to obtain features characterising both short- and long-term time dependencies in the data.

A general framework for sensor-based human activity recognition.

Today's wearable devices like smartphones, smartwatches and intelligent glasses collect a large amount of data from their built-in sensors like accelerometers and gyroscopes. These data can be used to identify a person's current activity and in turn can be utilised for applications in the field of personal fitness assistants or elderly care. However, developing such systems is subject to certain restrictions: (i) since more and more new sensors will be available in the future, activity recognition systems should be able to integrate these new sensors with a small amount of manual effort and (ii) such systems should avoid high acquisition costs for computational power. We propose a general framework that achieves an effective data integration based on the following two characteristics: Firstly, a smartphone is used to gather and temporally store data from different sensors and transfer these data to a central server. Thus, various sensors can be integrated into the system as long as they have programming interfaces to communicate with the smartphone. The second characteristic is a codebook-based feature learning approach that can encode data from each sensor into an effective feature vector only by tuning a few intuitive parameters. In the experiments, the framework is realised as a real-time activity recognition system that integrates eight sensors from a smartphone, smartwatch and smartglasses, and its effectiveness is validated from different perspectives such as accuracies, sensor combinations and sampling rates.