Summary of the sussex-huawei locomotion-transportation recognition challenge 2020

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

doi:10.1145/3410530.3414341

Cited by 42 publications

(25 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The problem of distribution gap between training and testing has been mentioned in the previous challenge [17] 3. The gap is large and Running, and Bus prediction has an awful result on the validation dataset.…”

Section: Training On Initial Training Datasetmentioning

confidence: 99%

“…SHL(short for Sussex-Huawei Locomotion) Recognition Challenge provides a phone sensor dataset for recognizing eight modes of locomotion and transportation activities, attracting worldwide researchers since 2018 [16,17,20]. Activities Locomotion can be regarded as a multi-class sequence labelling task whose target is a multi-class predicting label vector given a variable-length sequence.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Triple-O for SHL Recognition Challenge: An Ensemble Framework for Multi-class Imbalance and Training-testing Distribution Inconsistency by OvO Binarization with Confidence Weight of One-class Classification

Zhang

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

SHL Recognition Challenge provides phone sensor data for recognizing eight modes of locomotion and transportation(activities). Most motion sensor data is user-dependent, so the model's generalization ability requires various data at the user level. The SHL dataset is collected only by three users, and category imbalance and distribution gap between training data and testing data can be significant obstacles for this challenge. In comparison to exploring the large and deep model structure, some model-free tricks play a more critical role in the previous challenges, such as transfer learning, re-sampling and anti-overfit trick. SHL challenge also notices that problem and provides user-independent sensor data in 2021. This paper analyses the data and finds out that category imbalance and distribution inconsistency are still the obstacles. This paper focuses on evaluating different methods to improve the general predicting ability at the setting of category imbalance and training-testing distribution inconsistency. Besides, this paper puts forward a new ensemble framework called triple-O, using OvO binarization and one-class classification. The results show that the OvO binarization ensemble gets better results on the hard-to-distinguish class than re-sampling and re-weighting. One-class classification can be an anomaly detection to re-weight the meta learners to tackle the distribution gap. Triple-O can be plug-and-play and pave the way for exploring complex model structures. This paper introduces the solutions from the team GoodGoodDriveDayDayUp. CCS CONCEPTS• Computing methodologies → Bagging.

show abstract

Section: Training On Initial Training Datasetmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Triple-O for SHL Recognition Challenge: An Ensemble Framework for Multi-class Imbalance and Training-testing Distribution Inconsistency by OvO Binarization with Confidence Weight of One-class Classification

Zhang

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

“…In the previous three years of competition, data obtained from the use of motion sensors (accelerator and gyro, etc.) were used [24][20] [21]. Moreover, traditional machine learning and deep neural network models were proposed [23][7] to tackle the challenges arising from learning HAR models through the use of motion sensors.…”

Section: Background and Related Workmentioning

confidence: 99%

Classical machine learning and deep neural network ensemble model for GPS-based activity recognition

Kojima

Legaspi

Mishima

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

Our KDDI Research team proposes an ensemble model for the Sussex-Huawei Locomotion-Transportation (SHL) recognition challenge using GPS data. During preprocessing, we corrected the GPS dataset that contains errors and missing values using a Kalman smoother. Since this smoother can be processed offline, it can be used to correct backforward using the recorded future locations in the dataset. Further, by using features that have similar distributions in the train subject's data and the other subjects' datasets, we could achieve robust feature selection across multiple subjects. The first stage of our ensemble model employs classical machine learning and deep neural network approaches independently, specifically, LightGBM and LSTM, respectively. The second stage calculates the weighted average of the outputs of both approaches. Our results show the improved accuracy contributed by our ensemble, suggesting that it effectively makes use of both statistical and nonstatistical features given the suitable base models. We confirmed that the use of Kalman-smoother, selection of features with similar distributions across subjects and ensemble modeling contributed to improving the accuracy of both train and validation datasets.

show abstract

“…There are two types of activity recognition: sensor-based [13][14][15][16][17][18][19][20][21] and vision-based [10][11][12]. In recent years, sensor data have been used not only by the lower-level applications but also by the higher-level applications such as smart home, robot, and NLP applications [8].…”

Section: Introductionmentioning

confidence: 99%

Activity Knowledge Graph Recognition by Eye Gaze: Identification of Distant Object in Eye Sight for Watch Activity

Toyosaka

Okita

2021

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

Self Cite

View full text Add to dashboard Cite

From the side of upper-level applications which require planning the actions in robot or those which need to search the whole log of activities in smart home, the action predicate expressions in the form of knowledge graphs may play an important role. The sequence of activities alone, which can be supplied by the conventional activity recognition systems, may not be sufficient for those applications. The subject of the particular activity is crucial information in most of the cases, and the object of the particular activity is often necessary to identify the characteristics. From this perspective, we have investigated the activities recognized by activity recognition systems, trying to identify their hidden elements which play the role of the subject and the object of the activities, i.e. activity knowledge graph. If we focus on these hidden elements, they are categorized in two: (1) person (subject) -person (object) interaction, and (2) person (subject) -object (object) interactions. Depending on the class of activities, these two are sometimes faced great difficulties: the hidden elements for walk, pick-up, open, and drink are quite easy but those for look-at, see, watch, and throw are difficult. The source of difficulties arises from the fact that the object (object) is not contacted from the person (subject). In this paper we have developed a method which identifies non-contacted object by the direction of the eye gaze of the person (subject) in the category of watch (activity). Using "Watching TV" data by Stair lab, the proposed system achieved 85% in accuracy.

show abstract

Summary of the sussex-huawei locomotion-transportation recognition challenge 2020

Cited by 42 publications

References 34 publications

Triple-O for SHL Recognition Challenge: An Ensemble Framework for Multi-class Imbalance and Training-testing Distribution Inconsistency by OvO Binarization with Confidence Weight of One-class Classification

Triple-O for SHL Recognition Challenge: An Ensemble Framework for Multi-class Imbalance and Training-testing Distribution Inconsistency by OvO Binarization with Confidence Weight of One-class Classification

Classical machine learning and deep neural network ensemble model for GPS-based activity recognition

Activity Knowledge Graph Recognition by Eye Gaze: Identification of Distant Object in Eye Sight for Watch Activity

Contact Info

Product

Resources

About