pith. sign in

arxiv: 2606.24781 · v1 · pith:AMVSYISSnew · submitted 2026-06-23 · 💻 cs.AI · cs.HC

Assessing Distribution Shift in Human Activity Recognition for Domain Generalization

Rebecca Adaimi , Edison Thomaz This is my paper

Pith reviewed 2026-06-25 23:19 UTC · model grok-4.3

classification 💻 cs.AI cs.HC
keywords human activity recognitiondistribution shiftdomain generalizationsensor heterogeneitybenchmarkdevice variation
0
0 comments X

The pith

Diversity shifts define all distribution shifts in human activity recognition, revealing unique unshared features across domains.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper evaluates four types of distribution shifts in human activity recognition models, covering device type, sensor placement, sampling rate, and user behavior. It concludes that diversity shifts predominate in all cases, which means each domain carries unique features that do not appear in the others. The authors build a single benchmark for these shifts and test up to 28 domain generalization methods on it. Those methods improve only marginally over plain empirical risk minimization training. The results matter because real-world activity recognition must handle exactly these device and context differences to work outside controlled settings.

Core claim

Diversity shifts predominantly define all types of shifts, indicating the existence of unique features that are not shared across different domains. Domain generalization methods marginally outperform the empirical risk minimization baseline.

What carries the argument

A uniform HAR-based distribution shift benchmark that measures the four shift types and runs head-to-head tests of domain generalization algorithms against empirical risk minimization.

If this is right

  • Cross-domain performance in HAR cannot rely on features shared by all domains.
  • Standard training leaves performance gaps that current generalization techniques close only slightly.
  • Any deployed HAR system must account for device, placement, rate, and behavior differences from the start.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Real deployments will need either much larger and more varied training collections or runtime adaptation mechanisms beyond today's domain generalization methods.
  • The benchmark approach could be applied to other sensor-driven tasks such as fall detection or gesture interfaces to check whether diversity shifts dominate there as well.
  • Algorithm designers could target time-series sensor properties explicitly rather than importing image or text domain generalization techniques unchanged.

Load-bearing premise

The four chosen distribution shifts and the selected datasets are representative enough of real-world HAR variability that conclusions about diversity shifts and DG method performance will generalize beyond the tested cases.

What would settle it

Running the same evaluation protocol on a fresh collection of HAR datasets where non-diversity shifts predominate or where at least one domain generalization method shows large, consistent gains over the baseline.

Figures

Figures reproduced from arXiv: 2606.24781 by Edison Thomaz, Rebecca Adaimi.

Figure 1
Figure 1. Figure 1: Accelerometer signals captured with varying (a) sensor location (wrist, ankle, chest), (b) Device type [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Screenshots from video recorded sessions of a participant juggling 5 and 7 balls. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Domain Classifier model overview. The neural network extracts features from the accelerometer data [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
read the original abstract

While the field of Human Activity Recognition (HAR) continues to draw interest from researchers and advance in important ways, some key challenges remain. One of the most difficult aspects of building HAR models that show good performance in real-world settings is dealing with data diversity from device and sensor heterogeneity, and contextual changes that are intrinsic to real-world applications. While data diversity in HAR has been well-acknowledged in the literature, there remains a gap in understanding the effect of various types of distribution shifts on HAR models and the domain generalization problem that arises. Towards that end, this paper systematically evaluates 4 different types of distribution shifts, including variations in device type, sensor placement, sampling rate, and user behavior. Quantifying their effects, we illustrate that diversity shifts predominantly define all types of shifts, indicating the existence of unique features that are not shared across different domains. We then introduce a uniform HAR-based distribution shift benchmarks and conduct a comprehensive evaluation of up to 28 domain generalization methods. Our analysis exposes the limitations of current domain generalization algorithms in achieving model generalizability, marginally outperforming the empirical risk minimization baseline. This work represents the first systematic exploration of domain generalization and adaptation concerning specific distribution shifts in sensor-based HAR, offering an open-source benchmark platform and datasets to spur further research.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The manuscript systematically evaluates four types of distribution shifts in sensor-based Human Activity Recognition (device type, sensor placement, sampling rate, user behavior) across datasets. It concludes that diversity shifts predominate, indicating unique non-shared features across domains. The authors introduce a uniform benchmark and evaluate up to 28 domain generalization methods, reporting that they marginally outperform the empirical risk minimization baseline. The work positions itself as the first systematic study of DG/DA for specific shifts in HAR and releases an open-source benchmark platform and datasets.

Significance. If the empirical results hold, this provides a useful benchmark for the HAR community by quantifying shift impacts and exposing limitations of existing DG algorithms relative to ERM. The open-source release strengthens reproducibility and follow-on work. The representativeness concern (whether the four shifts and datasets capture real-world HAR variability) is an external-validity issue rather than an internal inconsistency in the argument; the paper's internal claims rest on the reported comparisons rather than untestable assumptions.

major comments (2)
  1. [§5 (Benchmark Evaluation)] §5 (Benchmark Evaluation) and associated results: the claim that 'diversity shifts predominantly define all types of shifts' is central but lacks an explicit quantitative metric or decision rule for determining predominance; without this, it is unclear how interactions between shift types were handled or whether the conclusion is robust to alternative aggregation methods.
  2. [DG method comparison tables/results] DG method comparison tables/results: the reported marginal outperformance over ERM is presented without statistical significance tests, confidence intervals, or multi-seed variance; this makes it impossible to determine whether the gap is reliable or could be explained by baseline implementation details or random variation.
minor comments (3)
  1. [Abstract] Abstract: 'up to 28' DG methods should be replaced with the exact count and a brief enumeration or reference to the full list for immediate clarity.
  2. [§3] Notation and terminology: consistent use of 'diversity shift' versus other shift categories would benefit from an early definitional table or paragraph.
  3. [Figures in §4-5] Figure captions and axis labels in the shift quantification plots could include explicit dataset and shift-type identifiers to improve standalone readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive recommendation. We address the two major comments below and will revise the manuscript to strengthen the presentation of results.

read point-by-point responses

  1. Referee: [§5 (Benchmark Evaluation)] §5 (Benchmark Evaluation) and associated results: the claim that 'diversity shifts predominantly define all types of shifts' is central but lacks an explicit quantitative metric or decision rule for determining predominance; without this, it is unclear how interactions between shift types were handled or whether the conclusion is robust to alternative aggregation methods.

    Authors: We agree that an explicit quantitative metric would improve clarity. In the revision we will add a formal predominance score defined as the fraction of total performance drop (relative to in-domain ERM) that is uniquely attributable to each shift type after subtracting shared components via a simple additive decomposition. We will also report results under two alternative aggregation schemes (weighted by dataset size and by shift severity) to demonstrate robustness. Interactions between shift types will be handled by including both single-shift and combined-shift experiments with the new metric applied to each. revision: yes

  2. Referee: [DG method comparison tables/results] DG method comparison tables/results: the reported marginal outperformance over ERM is presented without statistical significance tests, confidence intervals, or multi-seed variance; this makes it impossible to determine whether the gap is reliable or could be explained by baseline implementation details or random variation.

    Authors: We acknowledge this limitation. The revision will include results averaged over five random seeds with reported standard deviations, 95% confidence intervals, and paired t-tests (or Wilcoxon tests where normality assumptions fail) comparing each DG method against ERM on the same splits. This will allow readers to assess whether the observed marginal gains are statistically distinguishable from zero or from implementation noise. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical benchmark only

full rationale

The paper performs an empirical evaluation of distribution shifts and domain generalization methods on HAR datasets. No mathematical derivation, fitted parameters renamed as predictions, self-definitional constructs, or load-bearing self-citations appear in the abstract or described methodology. All claims rest on direct experimental comparisons (28 DG methods vs. ERM baseline across four shift types) that are externally falsifiable via the released benchmarks and datasets. The central finding that diversity shifts predominate is a measured outcome, not a reduction to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper is an empirical ML evaluation study. It relies on standard supervised learning assumptions and the representativeness of chosen datasets and shift types. No new entities are postulated and no free parameters are introduced beyond typical hyperparameter choices in the evaluated methods.

axioms (2)
  • domain assumption The four distribution shift types and chosen datasets capture the dominant sources of variability in real-world sensor-based HAR.
    This premise is required for the claim that diversity shifts predominate and that DG methods are generally limited.
  • standard math Standard i.i.d. assumptions within each domain and the validity of ERM as a baseline hold for the evaluation protocol.
    Implicit in all domain generalization benchmarking.

pith-pipeline@v0.9.1-grok · 5752 in / 1334 out tokens · 29203 ms · 2026-06-25T23:19:07.921873+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

71 extracted references · 16 canonical work pages · 3 internal anchors

  1. [1]

    Rebecca Adaimi, Howard Yong, and Edison Thomaz. 2021. Ok Google, What Am I Doing? Acoustic Activity Recognition Bounded by Conversational Assistant Interactions. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 5, 1 (2021), 1–24

    2021

  2. [2]

    Kartik Ahuja, Ethan Caballero, Dinghuai Zhang, Jean-Christophe Gagnon-Audet, Yoshua Bengio, Ioannis Mitliagkas, and Irina Rish. 2021. Invariance principle meets information bottleneck for out-of-distribution generalization.Advances in Neural Information Processing Systems 34 (2021), 3438–3450

    2021

  3. [3]

    Ali Akbari and Roozbeh Jafari. 2019. Transferring activity recognition models for new wearable sensors with deep generative domain adaptation. In Proceedings of the 18th International Conference on Information Processing in Sensor Networks. 85–96. Assessing Distribution Shift in Human Activity Recognition for Domain Generalization 19

    2019

  4. [4]

    Kerem Altun and Billur Barshan. 2010. Human activity recognition using inertial/magnetic sensor units. In Human Behavior Understanding: First International Workshop, HBU 2010, Istanbul, Turkey, August 22, 2010. Proceedings 1 . Springer, 38–51

    2010

  5. [5]

    Martin Arjovsky, Léon Bottou, Ishaan Gulrajani, and David Lopez-Paz. 2019. Invariant risk minimization. arXiv preprint arXiv:1907.02893 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  6. [6]

    Sarnab Bhattacharya, Rebecca Adaimi, and Edison Thomaz. 2022. Leveraging Sound and Wrist Motion to Detect Activities of Daily Living with Commodity Smartwatches. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 6, 2, Article 42 (jul 2022), 28 pages. https://doi.org/10.1145/3534582

    work page doi:10.1145/3534582 2022

  7. [7]

    Gilles Blanchard, Aniket Anand Deshmukh, Ürun Dogan, Gyemin Lee, and Clayton Scott. 2021. Domain generalization by marginal transfer learning. The Journal of Machine Learning Research 22, 1 (2021), 46–100

    2021

  8. [8]

    Hennie Brugman, Albert Russel, and Xd Nijmegen. 2004. Annotating Multi-media/Multi-modal Resources with ELAN.. In LREC. 2065–2068

    2004

  9. [9]

    Ayse S Cakmak, Nina Thigpen, Garrett Honke, Erick Perez Alday, Ali Bahrami Rad, Rebecca Adaimi, Chia Jung Chang, Qiao Li, Pramod Gupta, Thomas Neylan, et al. 2020. Using Convolutional Variational Autoencoders to Predict Post-Trauma Health Outcomes from Actigraphy Data. In Machine Learning for Mobile Health Workshop at NeurIPS

    2020

  10. [10]

    Avijoy Chakma, Abu Zaher Md Faridee, Md Abdullah Al Hafiz Khan, and Nirmalya Roy. 2021. Activity recognition in wearables using adversarial multi-source domain adaptation. Smart Health 19 (2021), 100174

    2021

  11. [11]

    Youngjae Chang, Akhil Mathur, Anton Isopoussu, Junehwa Song, and Fahim Kawsar. 2020. A systematic study of unsupervised domain adaptation for robust human-activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 4, 1 (2020), 1–30

    2020

  12. [12]

    Ricardo Chavarriaga, Hesam Sagha, Alberto Calatroni, Sundara Tejaswi Digumarti, Gerhard Tröster, José del R Millán, and Daniel Roggen. 2013. The Opportunity challenge: A benchmark database for on-body sensor-based activity recognition. Pattern Recognition Letters 34, 15 (2013), 2033–2042

    2013

  13. [13]

    Changhao Chen, Yishu Miao, Chris Xiaoxuan Lu, Linhai Xie, Phil Blunsom, Andrew Markham, and Niki Trigoni. 2019. Motiontransformer: Transferring neural inertial tracking between domains. In Proceedings of the AAAI conference on artificial intelligence, Vol. 33. 8009–8016

    2019

  14. [14]

    Mathieu Chevalley, Charlotte Bunne, Andreas Krause, and Stefan Bauer. 2022. Invariant causal mechanisms through distribution matching. arXiv preprint arXiv:2206.11646 (2022)

    work page arXiv 2022

  15. [15]

    Yi-Ting Chiang, Ching-Hu Lu, and Jane Yung-Jen Hsu. 2017. A Feature-Based Knowledge Transfer Framework for Cross-Environment Activity Recognition Toward Smart Home Applications. IEEE Transactions on Human-Machine Systems 47, 3 (2017), 310–322. https://doi.org/10.1109/THMS.2016.2641679

    work page doi:10.1109/thms.2016.2641679 2017

  16. [16]

    Sanders, Rebecca Adaimi, Necole Streeper, David E

    Keum San Chun, Ashley B. Sanders, Rebecca Adaimi, Necole Streeper, David E. Conroy, and Edison Thomaz. 2019. Towards a Generalizable Method for Detecting Fluid Intake with Wrist-Mounted Sensors and Adaptive Segmentation. In Proceedings of the 24th International Conference on Intelligent User Interfaces (Marina del Ray, California) (IUI ’19). Association f...

    work page doi:10.1145/3301275.3302315 2019

  17. [17]

    Cian Eastwood, Alexander Robey, Shashank Singh, Julius Von Kügelgen, Hamed Hassani, George J Pappas, and Bern- hard Schölkopf. 2022. Probable domain generalization via quantile risk minimization. arXiv preprint arXiv:2207.09944 (2022)

    work page arXiv 2022

  18. [18]

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, François Laviolette, Mario Marchand, and Victor Lempitsky. 2016. Domain-adversarial training of neural networks. The journal of machine learning research 17, 1 (2016), 2096–2030

    2016

  19. [19]

    Martin Gjoreski, Stefan Kalabakov, Mitja Luštrek, Matjaž Gams, and Hristijan Gjoreski. 2019. Cross-dataset deep transfer learning for activity recognition. In Adjunct Proceedings of the 2019 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2019 ACM International Symposium on Wearable Computers . 714–718

    2019

  20. [20]

    Ishaan Gulrajani and David Lopez-Paz. 2020. In search of lost domain generalization. arXiv preprint arXiv:2007.01434 (2020)

    work page arXiv 2020

  21. [21]

    Zeyi Huang, Haohan Wang, Eric P Xing, and Dong Huang. 2020. Self-challenging improves cross-domain generalization. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16 . Springer, 124–140

    2020

  22. [22]

    Naoto Inoue, Ryosuke Furuta, Toshihiko Yamasaki, and Kiyoharu Aizawa. 2018. Cross-domain weakly-supervised object detection through progressive domain adaptation. In Proceedings of the IEEE conference on computer vision and pattern recognition. 5001–5009

    2018

  23. [23]

    Md Abdullah Al Hafiz Khan, Nirmalya Roy, and Archan Misra. 2018. Scaling human activity recognition via deep learning-based domain adaptation. In 2018 IEEE international conference on pervasive computing and communications (PerCom). IEEE, 1–9. 20 Adaimi, et al

    2018

  24. [24]

    Daehee Kim, Youngjun Yoo, Seunghyun Park, Jinkyu Kim, and Jaekoo Lee. 2021. Selfreg: Self-supervised contrastive regularization for domain generalization. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 9619–9628

    2021

  25. [25]

    Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsubramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, et al. 2021. Wilds: A benchmark of in-the-wild distribution shifts. In International Conference on Machine Learning . PMLR, 5637–5664

    2021

  26. [26]

    Masanori Koyama and Shoichiro Yamaguchi. 2020. When is invariance useful in an Out-of-Distribution Generalization problem? arXiv preprint arXiv:2008.01883 (2020)

    work page arXiv 2020

  27. [27]

    David Krueger, Ethan Caballero, Joern-Henrik Jacobsen, Amy Zhang, Jonathan Binas, Dinghuai Zhang, Remi Le Priol, and Aaron Courville. 2021. Out-of-distribution generalization via risk extrapolation (rex). In International Conference on Machine Learning. PMLR, 5815–5826

    2021

  28. [28]

    Da Li, Yongxin Yang, Yi-Zhe Song, and Timothy Hospedales. 2018. Learning to generalize: Meta-learning for domain generalization. In Proceedings of the AAAI conference on artificial intelligence , Vol. 32

    2018

  29. [29]

    Haoliang Li, Sinno Jialin Pan, Shiqi Wang, and Alex C Kot. 2018. Domain generalization with adversarial feature learning. (2018), 5400–5409

    2018

  30. [30]

    Haoliang Li, YuFei Wang, Renjie Wan, Shiqi Wang, Tie-Qiang Li, and Alex Kot. 2020. Domain generalization for medical imaging classification with linear-dependency regularization. Advances in Neural Information Processing Systems 33 (2020), 3118–3129

    2020

  31. [31]

    Jing Li, Shuo Shang, and Lisi Chen. 2020. Domain generalization for named entity boundary detection via metalearning. IEEE Transactions on Neural Networks and Learning Systems 32, 9 (2020), 3819–3830

    2020

  32. [32]

    Pan Li, Da Li, Wei Li, Shaogang Gong, Yanwei Fu, and Timothy M Hospedales. 2021. A simple feature augmentation for domain generalization. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 8886–8895

    2021

  33. [33]

    Xinyu Li, Yuan He, J Andrew Zhang, and Xiaojun Jing. 2021. Supervised domain adaptation for few-shot radar-based human activity recognition. IEEE Sensors Journal 21, 22 (2021), 25880–25890

    2021

  34. [34]

    Ya Li, Xinmei Tian, Mingming Gong, Yajing Liu, Tongliang Liu, Kun Zhang, and Dacheng Tao. 2018. Deep domain generalization via conditional invariant adversarial networks. In Proceedings of the European conference on computer vision (ECCV). 624–639

    2018

  35. [35]

    Wang Lu, Jindong Wang, Yiqiang Chen, Sinno Jialin Pan, Chunyu Hu, and Xin Qin. 2022. Semantic-discriminative mixup for generalizable sensor-based cross-domain activity recognition. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 6, 2 (2022), 1–19

    2022

  36. [36]

    Francisco Javier Ordóñez Morales and Daniel Roggen. 2016. Deep Convolutional Feature Transfer across Mobile Activity Recognition Domains, Sensor Modalities and Locations. In Proceedings of the 2016 ACM International Symposium on Wearable Computers (Heidelberg, Germany) (ISWC ’16). Association for Computing Machinery, New York, NY, USA, 92–99. https://doi....

    work page doi:10.1145/2971763.2971764 2016

  37. [37]

    Hyeonseob Nam, HyunJae Lee, Jongchan Park, Wonjun Yoon, and Donggeun Yoo. 2021. Reducing domain gap by reducing style bias. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 8690–8699

    2021

  38. [38]

    Shirin Nilizadeh, Hojjat Aghakhani, Eric Gustafson, Christopher Kruegel, and Giovanni Vigna. 2019. Think outside the dataset: Finding fraudulent reviews using cross-dataset analysis. In The World Wide Web Conference. 3108–3115

    2019

  39. [39]

    Giambattista Parascandolo, Alexander Neitz, Antonio Orvieto, Luigi Gresele, and Bernhard Schölkopf. 2020. Learning explanations that are hard to vary. arXiv preprint arXiv:2009.00329 (2020)

    work page arXiv 2020

  40. [40]

    Mohammad Pezeshki, Oumar Kaba, Yoshua Bengio, Aaron C Courville, Doina Precup, and Guillaume Lajoie. 2021. Gradient starvation: A learning proclivity in neural networks. Advances in Neural Information Processing Systems 34 (2021), 1256–1272

    2021

  41. [41]

    Xin Qin, Jindong Wang, Yiqiang Chen, Wang Lu, and Xinlong Jiang. 2022. Domain Generalization for Activity Recognition via Adaptive Feature Fusion. ACM Transactions on Intelligent Systems and Technology 14, 1 (2022), 1–21

    2022

  42. [42]

    Mohamed Ragab, Emadeldeen Eldele, Wee Ling Tan, Chuan-Sheng Foo, Zhenghua Chen, Min Wu, Chee-Keong Kwoh, and Xiaoli Li. 2023. Adatime: A benchmarking suite for domain adaptation on time series data. ACM Transactions on Knowledge Discovery from Data 17, 8 (2023), 1–18

    2023

  43. [43]

    Sreenivasan Ramasamy Ramamurthy and Nirmalya Roy. 2018. Recent trends in machine learning for human activity recognition—A survey. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 8, 4 (2018), e1254

    2018

  44. [44]

    Attila Reiss and Didier Stricker. 2012. Creating and benchmarking a new dataset for physical activity monitoring. In Proceedings of the 5th International Conference on PErvasive Technologies Related to Assistive Environments . 1–8

    2012

  45. [45]

    Pappas, and Hamed Hassani

    Alexander Robey, George J. Pappas, and Hamed Hassani. 2021. Model-Based Domain Generalization. In Advances in Neural Information Processing Systems , M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan (Eds.), Vol. 34. Curran Associates, Inc., 20210–20229. https://proceedings.neurips.cc/paper_files/paper/2021/file/ a8f12d9486cbcc2fe...

    2021

  46. [46]

    Maddison

    Yangjun Ruan, Yann Dubois, and Chris J. Maddison. 2022. Optimal Representations for Covariate Shift. InInternational Conference on Learning Representations . https://openreview.net/forum?id=Rf58LPCwJj0

    2022

  47. [47]

    Shiori Sagawa, Pang Wei Koh, Tatsunori B Hashimoto, and Percy Liang. 2019. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization. arXiv preprint arXiv:1911.08731 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  48. [48]

    Keum San Chun, Hyoyoung Jeong, Rebecca Adaimi, and Edison Thomaz. 2020. Eating episode detection with jawbone- mounted inertial sensing. In 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). IEEE, 4361–4364

    2020

  49. [49]

    Soroosh Shahtalebi, Jean-Christophe Gagnon-Audet, Touraj Laleh, Mojtaba Faramarzi, Kartik Ahuja, and Irina Rish

  50. [50]

    arXiv preprint arXiv:2106.02266 (2021)

    Sand-mask: An enhanced gradient masking strategy for the discovery of invariances in domain generalization. arXiv preprint arXiv:2106.02266 (2021)

    work page arXiv 2021

  51. [51]

    Yuge Shi, Jeffrey Seely, Philip HS Torr, N Siddharth, Awni Hannun, Nicolas Usunier, and Gabriel Synnaeve. 2021. Gradient matching for domain generalization. arXiv preprint arXiv:2104.09937 (2021)

    work page arXiv 2021

  52. [52]

    Allan Stisen, Henrik Blunck, Sourav Bhattacharya, Thor Siiger Prentow, Mikkel Baun Kjærgaard, Anind Dey, Tobias Sonne, and Mads Møller Jensen. 2015. Smart devices are different: Assessing and mitigatingmobile sensing hetero- geneities for activity recognition. In Proceedings of the 13th ACM conference on embedded networked sensor systems . 127–140

    2015

  53. [53]

    Baochen Sun and Kate Saenko. 2016. Deep coral: Correlation alignment for deep domain adaptation. (2016), 443–450

    2016

  54. [54]

    Shiliang Sun, Honglei Shi, and Yuanbin Wu. 2015. A survey of multi-source domain adaptation. Information Fusion 24 (2015), 84–92

    2015

  55. [55]

    Timo Sztyler and Heiner Stuckenschmidt. 2016. On-body localization of wearable devices: An investigation of position-aware activity recognition. In 2016 IEEE International Conference on Pervasive Computing and Communications (PerCom). IEEE, 1–9

    2016

  56. [56]

    Bernd Thomas. 2022. SensorLog. App Store. https://apps. apple. com/us/app/sensorlog/id388014573 (accessed Nov. 13,

    2022

  57. [57]

    Eric Tzeng, Judy Hoffman, Trevor Darrell, and Kate Saenko. 2015. Simultaneous deep transfer across domains and tasks. In Proceedings of the IEEE international conference on computer vision . 4068–4076

    2015

  58. [58]

    Vladimir N Vapnik. 1999. An overview of statistical learning theory. IEEE transactions on neural networks 10, 5 (1999), 988–999

    1999

  59. [59]

    Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, Tao Qin, Wang Lu, Yiqiang Chen, Wenjun Zeng, and Philip Yu. 2022. Generalizing to unseen domains: A survey on domain generalization. IEEE Transactions on Knowledge and Data Engineering (2022)

    2022

  60. [60]

    Jindong Wang, Vincent W Zheng, Yiqiang Chen, and Meiyu Huang. 2018. Deep transfer learning for cross-domain activity recognition. In proceedings of the 3rd International Conference on Crowd Science and Engineering . 1–8

    2018

  61. [61]

    Shujun Wang, Lequan Yu, Caizi Li, Chi-Wing Fu, and Pheng-Ann Heng. 2020. Learning from extrinsic and intrinsic supervisions for domain generalization. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part IX . Springer, 159–176

    2020

  62. [62]

    Xuhai Xu, Xin Liu, Han Zhang, Weichen Wang, Subigya Nepal, Yasaman Sefidgar, Woosuk Seo, Kevin S Kuehn, Jeremy F Huckins, Margaret E Morris, et al. 2023. GLOBEM: Cross-Dataset Generalization of Longitudinal Human Behavior Modeling. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 6, 4 (2023), 1–34

    2023

  63. [63]

    Yilun Xu and Tommi Jaakkola. 2021. Learning representations that support robust transfer of predictors.arXiv preprint arXiv:2110.09940 (2021)

    work page arXiv 2021

  64. [64]

    Shen Yan, Huan Song, Nanxiang Li, Lincan Zou, and Liu Ren. 2020. Improve unsupervised domain adaptation with mixup training. arXiv preprint arXiv:2001.00677 (2020)

    work page arXiv 2020

  65. [65]

    Yi Yao and Gianfranco Doretto. 2010. Boosting for transfer learning with multiple sources. In 2010 IEEE computer society conference on computer vision and pattern recognition . IEEE, 1855–1862

    2010

  66. [66]

    Nanyang Ye, Kaican Li, Haoyue Bai, Runpeng Yu, Lanqing Hong, Fengwei Zhou, Zhenguo Li, and Jun Zhu. 2022. Ood-bench: Quantifying and understanding two dimensions of out-of-distribution generalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 7947–7958

    2022

  67. [67]

    Guojun Zhang, Han Zhao, Yaoliang Yu, and Pascal Poupart. 2021. Quantifying and improving transferability in domain generalization. Advances in Neural Information Processing Systems 34 (2021), 10957–10970

    2021

  68. [68]

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. 2017. mixup: Beyond empirical risk minimization. arXiv preprint arXiv:1710.09412 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  69. [69]

    Marvin Zhang, Henrik Marklund, Nikita Dhawan, Abhishek Gupta, Sergey Levine, and Chelsea Finn. 2021. Adaptive risk minimization: Learning to adapt to domain shift. Advances in Neural Information Processing Systems 34 (2021), 23664–23678. 22 Adaimi, et al

    2021

  70. [70]

    Jiachen Zhao, Fang Deng, Haibo He, and Jie Chen. 2020. Local domain adaptation for cross-domain activity recognition. IEEE Transactions on Human-Machine Systems 51, 1 (2020), 12–21

    2020

  71. [71]

    Kaiyang Zhou, Ziwei Liu, Yu Qiao, Tao Xiang, and Chen Change Loy. 2022. Domain generalization: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence (2022). Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009

    2022