Summary of the Sussex-Huawei locomotion-transportation recognition challenge 2019

Wang, Lin; Gjoreski, Hristijan; Ciliberto, Mathias; Lago, Paula; Murao, Kazuya; Okita, Tsuyoshi; Roggen, Daniel

doi:10.1145/3341162.3344872

Cited by 47 publications

(21 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unlike the previous installments of this challenge [18], [15], [16] this year, the data is based on radio sensors, including GPS reception, GPS location, Wi-Fi reception, and GSM cell tower scans. So, the increased number of data variety makes it more important to extract related features that are highly correlated to the target labels of this challenge.…”

Section: Feature Extractionmentioning

confidence: 99%

A Windowless Approach to Recognize Various Modes of Locomotion and Transportation

Basak

Tasin

Sakib

et al. 2021

Adjunct Proceedings of the 2021 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of The

View full text Add to dashboard Cite

Detecting modes of transportation through human activity recognition is important in the effective and smooth operation of smartphone applications or similar portable devices. However, the effectiveness of such tasks depends on the nature and type of data provided, and it can often become quite challenging. "SHL recognition challenge 2021" is an activity recognition challenge that aims to detect eight modes of locomotion and transportation. The dataset in this challenge was based on radio data, including GPS reception, GPS location, Wi-Fi reception, and GSM cell tower scans. The objective was to create a model that was able to recognize the modes in a user-independent manner. In this paper, our team (Team Nirban) gives an appropriate summarization of our methods and approach to the challenge. We processed the data, extracted various features from the dataset, and selected the best ones, which helped our model to be generative and user-independent. We exploited a classical machine learning based approach and achieved 93.4% accuracy and 89.6% F1 score on the training set using 10-fold crossvalidation, as well as 62.3% accuracy on the provided validation set. CCS CONCEPTS• Human-centered computing → Smartphones; • Computing methodologies → Supervised learning by classification; Classification and regression trees.

show abstract

Section: Feature Extractionmentioning

confidence: 99%

A Windowless Approach to Recognize Various Modes of Locomotion and Transportation

Basak

Tasin

Sakib

et al. 2021

Adjunct Proceedings of the 2021 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of The

View full text Add to dashboard Cite

show abstract

“…As a result, more people can be involved to decrease carbon emissions and positively affect the environment to minimize global warming issues. This year's challenge differs from previous versions [27,29,30] of the SHL challenge, which primarily focused on transportation mode recognition from the motion sensors. Radio data can enhance recognition possibilities of the models that will compound it with motion sensors information.…”

mentioning

confidence: 99%

Human activity recognition with AutoML using smartphone radio data

Balabka

Shkliarenko²

2021

Adjunct Proceedings of the 2021 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of The

View full text Add to dashboard Cite

Participants of the fourth edition of SHL recognition challenge 2021 aim to recognize eight locomotion and transportation activities in a user-independent manner based on radio data, including GPS reception, GPS location, WiFi reception, and GSM cell tower scans. Team "DD" proposes applying Google's AutoML Tables service to preprocess data, train, and evaluate the model. During this challenge, we showed the advantages and disadvantages of AutoML Tables. In addition, we have employed additional complimentary publically available datasets. AutoML Tables helped to train an artificial neural network model using the AdaNet algorithm. As a result, it has shown an ability to recognize classes with a precision of 81.2% and recall of 78.2%. Also, we opened the source code of the required feature engineering and published it on GitHub: https://github.com/dbalabka/shl-activity-recognition-2021. CCS CONCEPTS• Human-centered computing → Ubiquitous and mobile computing; • Computing methodologies → Machine learning; Machine learning approaches; Neural networks.

show abstract

“…Despite HAR Task being well investigated, attempts to benchmark it on smartphones are only recent. As reported in a recent survey [28], a large number of datasets acquired from smartphones, worn in different ways, with various sensors and sampling frequency, make it difficult to reach a uniformity in tasks, sensors, protocols, time windows, etc.…”

Section: A Har Task and Datasetsmentioning

confidence: 99%

A Federated Learning Aggregation Algorithm for Pervasive Computing: Evaluation and Comparison

Portet

Lalanda

et al. 2021

2021 IEEE International Conference on Pervasive Computing and Communications (PerCom)

View full text Add to dashboard Cite

Pervasive computing promotes the installation of connected devices in our living spaces in order to provide services. Two major developments have gained significant momentum recently: an advanced use of edge resources and the integration of machine learning techniques for engineering applications. This evolution raises major challenges, in particular related to the appropriate distribution of computing elements along an edgeto-cloud continuum. About this, Federated Learning has been recently proposed for distributed model training in the edge. The principle of this approach is to aggregate models learned on distributed clients in order to obtain a new, more general model. The resulting model is then redistributed to clients for further training. To date, the most popular federated learning algorithm uses coordinate-wise averaging of the model parameters for aggregation. However, it has been shown that this method is not adapted in heterogeneous environments where data is not identically and independently distributed (non-iid). This corresponds directly to some pervasive computing scenarios where heterogeneity of devices and users challenges machine learning with the double objective of generalization and personalization. In this paper, we propose a novel aggregation algorithm, termed FedDist, which is able to modify its model architecture (here, deep neural network) by identifying dissimilarities between specific neurons amongst the clients. This permits to account for clients' specificity without impairing generalization. Furthermore, we define a complete method to evaluate federated learning in a realistic way taking generalization and personalization into account.Using this method, FedDist is extensively tested and compared with three state-of-the-art federated learning algorithms on the pervasive domain of Human Activity Recognition with smartphones.

show abstract

Summary of the Sussex-Huawei locomotion-transportation recognition challenge 2019

Cited by 47 publications

References 23 publications

A Windowless Approach to Recognize Various Modes of Locomotion and Transportation

A Windowless Approach to Recognize Various Modes of Locomotion and Transportation

Human activity recognition with AutoML using smartphone radio data

A Federated Learning Aggregation Algorithm for Pervasive Computing: Evaluation and Comparison

Contact Info

Product

Resources

About