Robust Feature Representation Using Multi-Task Learning for Human Activity Recognition.

Learning underlying patterns from sensory data is crucial in the Human Activity Recognition (HAR) task to avoid poor generalization when coping with unseen data. A key solution to such an issue is representation learning, which becomes essential when input signals contain activities with similar patterns or when patterns generated by different subjects for the same activity vary. To address these issues, we seek a solution to increase generalization by learning the underlying factors of each sensor signal. We develop a novel multi-channel asymmetric auto-encoder to recreate input signals precisely and extract indicative unsupervised futures. Further, we investigate the role of various activation functions in signal reconstruction to ensure the model preserves the patterns of each activity in the output. Our main contribution is that we propose a multi-task learning model to enhance representation learning through shared layers between signal reconstruction and the HAR task to improve the robustness of the model in coping with users not included in the training phase. The proposed model learns shared features between different tasks that are indeed the underlying factors of each input signal. We validate our multi-task learning model using several publicly available HAR datasets, UCI-HAR, MHealth, PAMAP2, and USC-HAD, and an in-house alpine skiing dataset collected in the wild, where our model achieved 99%, 99%, 95%, 88%, and 92% accuracy. Our proposed method shows consistent performance and good generalization on all the datasets compared to the state of the art.

Counting Activities Using Weakly Labeled Raw Acceleration Data: A Variable-Length Sequence Approach with Deep Learning to Maintain Event Duration Flexibility.

This paper presents a novel approach for counting hand-performed activities using deep learning and inertial measurement units (IMUs). The particular challenge in this task is finding the correct window size for capturing activities with different durations. Traditionally, fixed window sizes have been used, which occasionally result in incorrectly represented activities. To address this limitation, we propose segmenting the time series data into variable-length sequences using ragged tensors to store and process the data. Additionally, our approach utilizes weakly labeled data to simplify the annotation process and reduce the time to prepare annotated data for machine learning algorithms. Thus, the model receives only partial information about the performed activity. Therefore, we propose an LSTM-based architecture, which takes into account both the ragged tensors and the weak labels. To the best of our knowledge, no prior studies attempted counting utilizing variable-size IMU acceleration data with relatively low computational requirements using the number of completed repetitions of hand-performed activities as a label. Hence, we present the data segmentation method we employed and the model architecture that we implemented to show the effectiveness of our approach. Our results are evaluated using the Skoda public dataset for Human activity recognition (HAR) and demonstrate a repetition error of ±1 even in the most challenging cases. The findings of this study have applications and can be beneficial for various fields, including healthcare, sports and fitness, human-computer interaction, robotics, and the manufacturing industry.

Alpine Skiing Activity Recognition Using Smartphone's IMUs.

Many studies on alpine skiing are limited to a few gates or collected data in controlled conditions. In contrast, it is more functional to have a sensor setup and a fast algorithm that can work in any situation, collect data, and distinguish alpine skiing activities for further analysis. This study aims to detect alpine skiing activities via smartphone inertial measurement units (IMU) in an unsupervised manner that is feasible for daily use. Data of full skiing sessions from novice to expert skiers were collected in varied conditions using smartphone IMU. The recorded data is preprocessed and analyzed using unsupervised algorithms to distinguish skiing activities from the other possible activities during a day of skiing. We employed a windowing strategy to extract features from different combinations of window size and sliding rate. To reduce the dimensionality of extracted features, we used Principal Component Analysis. Three unsupervised techniques were examined and compared: KMeans, Ward's methods, and Gaussian Mixture Model. The results show that unsupervised learning can detect alpine skiing activities accurately independent of skiers' skill level in any condition. Among the studied methods and settings, the best model had 99.25% accuracy.

On the effectiveness of multi-feature evacuation systems: an agent-based exploratory simulation study.

Evacuation modeling and simulation are usually used to explore different possibilities for evacuation, however, it is a real challenge to integrate different categories of characteristics in unified modeling space. In this paper, we propose an agent-based model of an evacuating crowd so that a comparative analysis of a different sets of parameters categorized as individual, social and technological aspects, is made possible. In particular, we focus on the question of rationality vs. emotionalism of individuals in a localized social context. In addition to that, we propose and model the concept of extended social influence, thereby embedding technological influence within the social influence, and analyze its impact on the efficiency of evacuation. NetLogo is used for simulating different variations in environments, evacuation strategies, and agents demographics. Simulation results revealed that there is no substantial advantage of informational overload on people, as this might work only in those situations, where there are fewer chances of herding. In more serious situations, people should be left alone to decide. They, however, could be trained in drills, to avoid panicking in such situations and concentrate on making their decisions solely based on the dynamics of their surroundings. It was also learned that distant connectivity has no apparent advantage and can be ruled out while designing an evacuation strategy based on these recommendations.

The Effect of Expertise on Gaze Behaviour in Laparoscopic Cholecystectomy.

Laparoscopic skills vary with experience and training of surgeons. The complexity of laparoscopic surgeries affects the cognitive resources of surgeons significantly and leads to many biliary injuries during surgeries. Assuming that experts are more focused, we investigated how the skill level of surgeons during live surgery is reflected through eye metrics. Throughout the study, we used five eye movement metrics classified under saccadic, fixations and pupillary metrics. Forty-two laparoscopic surgeries have been conducted with four surgeons belonging to three expertise levels (novice, semi-expert and expert) from which thirty-eight surgeries were considered in the study. With the use of mean, standard deviation and ANOVA test we found three reliable metrics which we can use to differentiate the skill levels during live surgeries.

A survey of context recognition in surgery.

With the introduction of operating rooms of the future context awareness has gained importance in the surgical environment. This paper organizes and reviews different approaches for recognition of context in surgery. Major electronic research databases were queried to obtain relevant publications submitted between the years 2010 and 2015. Three different types of context were identified: (i) the surgical workflow context, (ii) surgeon's cognitive and (iii) technical state context. A total of 52 relevant studies were identified and grouped based on the type of context detected and sensors used. Different approaches were summarized to provide recommendations for future research. There is still room for improvement in terms of methods used and evaluations performed. Machine learning should be used more extensively to uncover hidden relationships between different properties of the surgeon's state, particularly when performing cognitive context recognition. Furthermore, validation protocols should be improved by performing more evaluations in situ and with a higher number of unique participants. The paper also provides a structured outline of recent context recognition methods to facilitate development of new generation context-aware surgical support systems.

Human computer confluence applied in healthcare and rehabilitation.

Human computer confluence (HCC) is an ambitious research program studying how the emerging symbiotic relation between humans and computing devices can enable radically new forms of sensing, perception, interaction, and understanding. It is an interdisciplinary field, bringing together researches from horizons as various as pervasive computing, bio-signals processing, neuroscience, electronics, robotics, virtual & augmented reality, and provides an amazing potential for applications in medicine and rehabilitation.