mlsanityData to 7 Oct 2026

GenHAR: Generalizing Cross-domain Human Activity Recognition for Last-mile Delivery

arXiv 2605.22086 ↗First posted 21 May 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Human Activity Recognition (HAR) has shown remarkable effectiveness in various applications, such as smart healthcare and intelligent manufacturing. However, a major challenge faced by HAR is the distribution shift across different sensor data domains, which often leads to decreased performance when deployed for real-world applications. To address this issue, this paper introduces GenHAR, a novel framework designed to mitigate the domain gap by learning domain-invariant sensor representations. GenHAR aims to enhance the generalization capabilities of HAR on target domains purely with data from the source domain. The key novelty of GenHAR lies in two aspects. Firstly, GenHAR tokenizes sensor data and learns correlations among frequency sensor channel dimensions to improve the robustness of HAR models. Secondly, GenHAR improves the efficiency via selective masking and an efficient attention mechanism. We conduct a systematic analysis of GenHAR by comparing it with state-of-the-art HAR methods on real-world human activity datasets. Results show that GenHAR outperforms state-of-the-art methods by 9.97% in accuracy, and reduces Floating Point Operations by 6.4 times. Moreover, we deploy GenHAR at a leading logistics company in 4 cities, and have detected 2.15 billion real-time activities. We release our code at: https://github.com/Sensor-FoundationModel/GenHAR.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 39 of its 82 references that are in the map. First posted on arXiv 21 May 2026.

Builds on

39 of 82 references.

References · 39

  1. 2017Attention Is All You Needcited by 59,932
  2. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  3. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  4. 2016Layer Normalizationcited by 5,579
  5. 2022A Time Series is Worth 64 Words: Long-term Forecasting with Transformerscited by 1,981
  6. 2021Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecastingcited by 1,654
  7. 2021MLP-Mixer: An all-MLP Architecture for Visioncited by 1,441
  8. 2018Deep learning for time series classification: a reviewcited by 575
  9. 2016Time Series Classification from Scratch with Deep Neural Networks: A Strong Baselinecited by 358
  10. 2019High Frequency Component Helps Explain the Generalization of Convolutional Neural Networkscited by 310
  11. 2022Self-Supervised Contrastive Pre-Training For Time Series via Time-Frequency Consistencycited by 246
  12. 2021SelfReg: Self-supervised Contrastive Regularization for Domain Generalizationcited by 189
27 more
  1. 2020Deep Learning for Sensor-based Human Activity Recognition: Overview, Challenges and Opportunitiescited by 125
  2. 2019Multi-task Self-Supervised Learning for Human Activity Detectioncited by 98
  3. 2016DeepSense: A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processingcited by 84
  4. 2018Mobile Sensor Data Anonymizationcited by 70
  5. 2022A Broad Study of Pre-training for Domain Generalization and Adaptationcited by 64
  6. 2022Assessing the State of Self-Supervised Human Activity Recognition using Wearablescited by 63
  7. 2023Domain Adaptation for Time Series Under Feature and Label Shiftscited by 60
  8. 2021Gated Transformer Networks for Multivariate Time Series Classificationcited by 58
  9. 2020Contrastive Predictive Coding for Human Activity Recognitioncited by 54
  10. 2022Deep Frequency Filtering for Domain Generalizationcited by 41
  11. 2023A Survey on Deep Learning based Time Series Analysis with Frequency Transformationcited by 40
  12. 2022Domain-invariant Feature Exploration for Domain Generalizationcited by 34
  13. 2021SelfHAR: Improving Human Activity Recognition through Self-training with Unlabeled Datacited by 31
  14. 2022Semantic-Discriminative Mixup for Generalizable Sensor-based Cross-domain Activity Recognitioncited by 28
  15. 2018Deep Transfer Learning for Cross-domain Activity Recognitioncited by 25
  16. 2022ERNIE-GeoL: A Geography-and-Language Pre-trained Model and its Applications in Baidu Mapscited by 25
  17. 2022ColloSSL: Collaborative Self-Supervised Learning for Human Activity Recognitioncited by 22
  18. 2022SWL-Adapt: An Unsupervised Domain Adaptation Model with Sample Weight Learning for Cross-User Wearable Human Activity Recognitioncited by 21
  19. 2020Attend And Discriminate: Beyond the State-of-the-Art for Human Activity Recognition using Wearable Sensorscited by 19
  20. 2023Finding Order in Chaos: A Novel Data Augmentation Method for Time Series in Contrastive Learningcited by 18
  21. 2020Incremental Real-Time Personalization in Human Activity Recognition Using Domain Adaptive Batch Normalizationcited by 13
  22. 2023Generalizable Low-Resource Activity Recognition with Diverse and Discriminative Representation Learningcited by 12
  23. 2023Weakly Supervised Multi-Task Representation Learning for Human Activity Analysis Using Wearablescited by 10
  24. 2022Learning Disentangled Behaviour Patterns for Wearable-based Human Activity Recognitioncited by 9
  25. 2023Combining Public Human Activity Recognition Datasets to Mitigate Labeled Data Scarcitycited by 9
  26. 2021Predicting COVID-19 Spread from Large-Scale Mobility Datacited by 7
  27. 2023ConvBoost: Boosting ConvNets for Sensor-based Activity Recognitioncited by 3

Built on by

Not cited yet.