pith. machine review for the scientific record. sign in

arxiv: 2605.01295 · v1 · submitted 2026-05-02 · 💻 cs.LG · cs.AI

Recognition: unknown

Autonomous Drift Learning in Data Streams: A Unified Perspective

Xiaoyu Yang , En Yu , Jie Lu
Authors on Pith no claims yet

Pith reviewed 2026-05-09 14:42 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords concept driftdata streamsautonomous learningtaxonomynon-stationaritycontinual learningdrift adaptationtemporal generalization
0
0 comments X

The pith

A three-dimensional taxonomy classifies non-stationarity in autonomous learning into time, data, and model stream drifts.

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

The paper argues that the assumption of stationarity fails in autonomous learning systems and that focusing only on temporal shifts, as in traditional concept drift, is insufficient. It proposes a taxonomy with three dimensions tied to the system's operational state: time stream drift for stochastic versus rhythmic patterns, data stream drift for representation changes versus semantic shifts, and model stream drift for internal divergences such as plasticity, heterogeneity, and policy instability. This unification of existing work across 193 studies bridges concept drift adaptation, continual learning, and temporal generalization, outlining how to build systems that adapt through ongoing change.

Core claim

The paper establishes a three-dimensional taxonomy for drift in autonomous data streams. Time stream drift distinguishes arbitrary stochastic patterns from structural rhythmic dynamics. Data stream drift separates shifts in feature representations from changes in underlying semantics. Model stream drift characterizes endogenous divergence through sequential plasticity, decentralized heterogeneity, and policy instability. Applied to a review of 193 studies, the taxonomy unifies fragmented research paradigms and identifies open challenges toward self-evolving intelligent systems.

What carries the argument

The three-dimensional taxonomy of time stream drift, data stream drift, and model stream drift, organized by the operational state of the learning system.

If this is right

  • Methods from concept drift, continual learning, and temporal generalization can be reclassified and compared systematically across the three dimensions.
  • Future algorithms for autonomous systems should monitor and respond to drifts in multiple streams at once rather than addressing one type in isolation.
  • Open challenges concentrate on model stream drift, particularly policy instability and decentralized heterogeneity.
  • The taxonomy supplies a concrete roadmap for constructing systems that evolve autonomously amid continuous change.

Where Pith is reading between the lines

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

  • The taxonomy could be tested by applying it to label all drifts observed in a deployed autonomous agent over months of operation and checking for uncovered cases.
  • It suggests designing integrated monitoring that tracks time, data, and model streams concurrently, rather than separate detection modules.
  • The framework may help identify why certain adaptation techniques fail when deployed in long-running systems that accumulate multiple drift types.

Load-bearing premise

The three dimensions capture every relevant form of non-stationarity in autonomous learning systems without significant overlap or omission.

What would settle it

Discovery of a type of non-stationarity arising in an autonomous learning system that fits none of the three categories, or that clearly requires a fourth independent dimension, would falsify the taxonomy's completeness.

Figures

Figures reproduced from arXiv: 2605.01295 by En Yu, Jie Lu, Xiaoyu Yang.

Figure 1
Figure 1. Figure 1: An overview of the proposed drift learning framework. This framework categorizes drift phenomena into three view at source ↗
Figure 2
Figure 2. Figure 2: The hierarchical taxonomy structure of Drift Learning proposed in this survey. We categorize the field along three axes: view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison of the two sub-categories of time view at source ↗
Figure 4
Figure 4. Figure 4: Geometric illustration of data stream drift. (a) Original view at source ↗
Figure 5
Figure 5. Figure 5: The Stability-Plasticity Balance in Sequence Drift. view at source ↗
read the original abstract

In the pursuit of autonomous learning systems, the foundational assumption of stationarity, the premise that data distributions and model behaviors remain constant, is fundamentally untenable. Historically, the research community has addressed non-stationary environments almost exclusively under the scope of concept drift, focusing primarily on temporal shifts in streams. However, as learning systems become increasingly autonomous and complex, merely adapting to temporal non-stationarity is no longer sufficient. Evolving beyond this traditional perspective, we propose a novel, three-dimensional taxonomy that systematizes the field based on the operational state of the system. First, time stream drift distinguishes between stochastic arbitrary patterns and structural rhythmic dynamics. Second, data stream drift disentangles shifts in feature representations, identified as representation drift, from changes in underlying semantics, recognized as semantic drift. Third, model stream drift characterizes the internal endogenous divergence of learning systems through the lenses of sequential plasticity, decentralized heterogeneity, and policy instability. Based on this framework, we systematically review 193 representative studies and identify key open challenges. By bridging the fragmented paradigms of drift adaptation, continual learning, and temporal generalization, this survey outlines a roadmap for building self-evolving intelligent systems capable of learning autonomously through continuous change.

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 / 2 minor

Summary. The paper proposes a novel three-dimensional taxonomy for non-stationarity in autonomous learning systems, consisting of time stream drift (stochastic arbitrary patterns vs. structural rhythmic dynamics), data stream drift (representation drift vs. semantic drift), and model stream drift (sequential plasticity, decentralized heterogeneity, and policy instability). It applies this framework to systematically review 193 representative studies, identifies key open challenges, and positions the taxonomy as a bridge across concept drift adaptation, continual learning, and temporal generalization to support self-evolving intelligent systems.

Significance. If the taxonomy can be shown to organize the literature without substantial overlap or omission, the work would provide a useful unifying lens for fragmented research areas in non-stationary machine learning, potentially helping researchers identify gaps in autonomous adaptation and guiding development of more robust self-evolving systems.

major comments (2)
  1. [Abstract; taxonomy definition] Abstract and taxonomy definition section: the central claim that the three dimensions capture distinct operational states 'without significant overlap' is load-bearing for the systematization contribution, yet the description treats model stream drift as endogenous divergence while data stream drift includes semantic shifts that commonly induce policy instability and plasticity changes; no explicit disambiguation rules, decision criteria, or interaction analysis are provided for assigning the 193 studies to dimensions.
  2. [Review of 193 studies] Review methodology (presumably the section detailing the 193-study categorization): without documented rules for handling cases where semantic drift in the data stream induces model stream effects (e.g., policy instability), the non-overlap premise cannot be verified, weakening the claim that the taxonomy comprehensively systematizes the field.
minor comments (2)
  1. [Abstract] The abstract uses 'time stream drift' and 'data stream drift' terminology that could be clarified with a short table contrasting the three dimensions and their subcategories for reader orientation.
  2. [Review section] Ensure all 193 studies are referenced with explicit dimension assignments in a supplementary table or appendix to allow reproducibility of the categorization.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below, clarifying the taxonomy's design and indicating where revisions will strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract; taxonomy definition] Abstract and taxonomy definition section: the central claim that the three dimensions capture distinct operational states 'without significant overlap' is load-bearing for the systematization contribution, yet the description treats model stream drift as endogenous divergence while data stream drift includes semantic shifts that commonly induce policy instability and plasticity changes; no explicit disambiguation rules, decision criteria, or interaction analysis are provided for assigning the 193 studies to dimensions.

    Authors: The taxonomy classifies drifts by their primary operational locus within the autonomous system rather than by downstream effects: time stream addresses temporal patterns in the data arrival process, data stream captures observable changes in the input data itself (representation versus semantic), and model stream isolates endogenous model-internal dynamics (plasticity, heterogeneity, policy instability) that arise independently of direct data observation. Semantic drift is therefore assigned to the data stream dimension when the initiating change is in data semantics, even if it later affects model behavior; the model stream dimension is reserved for cases where the primary phenomenon is internal model divergence not directly traceable to a data shift. While the manuscript's abstract and definition section emphasize this distinction, we acknowledge that explicit decision criteria and interaction analysis were not detailed. In the revision we will expand the taxonomy definition section with a dedicated subsection providing disambiguation rules, decision trees for boundary cases, and discussion of cross-dimensional interactions to support verification of the 193-study assignments. revision: yes

  2. Referee: [Review of 193 studies] Review methodology (presumably the section detailing the 193-study categorization): without documented rules for handling cases where semantic drift in the data stream induces model stream effects (e.g., policy instability), the non-overlap premise cannot be verified, weakening the claim that the taxonomy comprehensively systematizes the field.

    Authors: The review methodology section outlines the literature search, inclusion criteria, and high-level categorization process used to select and assign the 193 studies. Studies were placed according to the primary focus of each work (e.g., papers whose core contribution targets data-semantic adaptation are categorized under data stream drift). We recognize that the absence of explicit protocols for induced cross-effects limits independent verification of the non-overlap claim. The revision will therefore add a new subsection to the review methodology that documents the full categorization protocol, including explicit rules for handling induced model-stream effects from data-stream drifts, along with examples from the 193 studies. This addition will allow readers to assess the systematization claim directly. revision: yes

Circularity Check

0 steps flagged

No circularity: taxonomy is a conceptual synthesis from external literature

full rationale

The paper is a survey proposing a three-dimensional taxonomy (time stream drift, data stream drift, model stream drift) to systematize non-stationarity in autonomous learning. It reviews 193 studies and outlines challenges, but advances no mathematical derivations, equations, fitted parameters, or predictions. The taxonomy dimensions are defined descriptively from operational states synthesized across prior external work, without self-referential definitions, fitted-input predictions, or load-bearing self-citation chains. No step reduces by construction to its own inputs; the framework is an organizational proposal, not a derived result.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that existing paradigms are fragmented and insufficient for autonomous systems, requiring a new unified taxonomy. No free parameters or invented entities are introduced.

axioms (1)
  • domain assumption The foundational assumption of stationarity in data distributions and model behaviors is untenable for autonomous learning systems.
    This is explicitly stated as the starting premise in the abstract.

pith-pipeline@v0.9.0 · 5502 in / 1306 out tokens · 43120 ms · 2026-05-09T14:42:22.368298+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

192 extracted references · 12 canonical work pages · 3 internal anchors

  1. [1]

    A survey on concept drift adaptation,

    J. Gama, I. ˇZliobait˙e, A. Bifet, M. Pechenizkiy, and A. Bouchachia, “A survey on concept drift adaptation,”ACM computing surveys (CSUR), vol. 46, no. 4, pp. 1–37, 2014

    2014

  2. [2]

    Learning under concept drift: A review,

    J. Lu, A. Liu, F. Dong, F. Gu, J. Gama, and G. Zhang, “Learning under concept drift: A review,”IEEE Transactions on Knowledge and Data Engineering, vol. 31, no. 12, pp. 2346–2363, 2018

    2018

  3. [3]

    DYSON: Dynamic Feature Space Self-Organization for Online Task-Free Class Incremental Learning,

    Y . He, Y . Chen, Y . Jin, S. Dong, X. Wei, and Y . Gong, “DYSON: Dynamic Feature Space Self-Organization for Online Task-Free Class Incremental Learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 23 741–23 751

    2024

  4. [4]

    Learning robust spectral dynamics for temporal domain generalization,

    E. Yu, J. Lu, X. Yang, G. Zhang, and Z. Fang, “Learning robust spectral dynamics for temporal domain generalization,” inThe Thirty- ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

  5. [5]

    DriftShield: Autonomous Fraud Detection via Actor-Critic Reinforcement Learning With Dynamic Feature Reweighting,

    J. Cao, W. Zheng, Y . Ge, and J. Wang, “DriftShield: Autonomous Fraud Detection via Actor-Critic Reinforcement Learning With Dynamic Feature Reweighting,”IEEE Open Journal of the Computer Society, vol. 6, pp. 1166–1177, 2025

    2025

  6. [6]

    MOS: Model Surgery for Pre-Trained Model-Based Class-Incremental Learning,

    H.-L. Sun, D.-W. Zhou, H. Zhao, L. Gan, D.-C. Zhan, and H.-J. Ye, “MOS: Model Surgery for Pre-Trained Model-Based Class-Incremental Learning,”Proceedings of the AAAI Conference on Artificial Intelli- gence, vol. 39, no. 19, pp. 20 699–20 707, Apr. 2025

    2025

  7. [7]

    Learning under Concept Drift: A Review,

    J. Lu, A. Liu, F. Dong, F. Gu, J. Gama, and G. Zhang, “Learning under Concept Drift: A Review,”IEEE Transactions on Knowledge and Data Engineering, vol. 31, no. 12, pp. 2346–2363, Dec. 2019

    2019

  8. [8]

    A Survey on Concept Drift in Process Mining,

    D. M. V . Sato, S. C. De Freitas, J. P. Barddal, and E. E. Scalabrin, “A Survey on Concept Drift in Process Mining,”ACM Comput. Surv., vol. 54, no. 9, pp. 189:1–189:38, Oct. 2021

    2021

  9. [9]

    Concept Drift Detection in Data Stream Mining : A literature review,

    S. Agrahari and A. K. Singh, “Concept Drift Detection in Data Stream Mining : A literature review,”Journal of King Saud University - Computer and Information Sciences, vol. 34, no. 10, Part B, pp. 9523– 9540, Nov. 2022

    2022

  10. [10]

    Learning Under Concept Drift for Regression—A Systematic Literature Review,

    M. Lima, M. Neto, T. S. Filho, and R. A. de A. Fagundes, “Learning Under Concept Drift for Regression—A Systematic Literature Review,” IEEE Access, vol. 10, pp. 45 410–45 429, 2022

    2022

  11. [11]

    A survey of active and passive concept drift handling methods,

    M. Han, Z. Chen, M. Li, H. Wu, and X. Zhang, “A survey of active and passive concept drift handling methods,”Computational Intelligence, vol. 38, no. 4, pp. 1492–1535, 2022

    2022

  12. [12]

    From concept drift to model degradation: An overview on performance-aware drift detectors,

    F. Bayram, B. S. Ahmed, and A. Kassler, “From concept drift to model degradation: An overview on performance-aware drift detectors,” Knowledge-Based Systems, vol. 245, p. 108632, Jun. 2022

    2022

  13. [13]

    Learning with drift detection,

    J. Gama, P. Medas, G. Castillo, and P. Rodrigues, “Learning with drift detection,” inBrazilian Symposium on Artificial Intelligence. Springer, 2004, pp. 286–295

    2004

  14. [14]

    Early drift detection method,

    M. Baena-Garcıa, J. del Campo- ´Avila, R. Fidalgo, A. Bifet, R. Gavalda, and R. Morales-Bueno, “Early drift detection method,” inFourth international workshop on knowledge discovery from data streams, vol. 6, 2006, pp. 77–86

    2006

  15. [15]

    Online and non- parametric drift detection methods based on hoeffding’s bounds,

    I. Frias-Blanco, J. del Campo- ´Avila, G. Ramos-Jimenez, R. Morales- Bueno, A. Ortiz-Diaz, and Y . Caballero-Mota, “Online and non- parametric drift detection methods based on hoeffding’s bounds,”IEEE Transactions on Knowledge and Data Engineering, vol. 27, no. 3, pp. 810–823, 2014

    2014

  16. [16]

    Learning from time-changing data with adaptive windowing,

    A. Bifet and R. Gavalda, “Learning from time-changing data with adaptive windowing,” inProceedings of the 2007 SIAM International Conference on Data Mining. SIAM, 2007, pp. 443–448

    2007

  17. [17]

    Early concept drift detection via prediction uncertainty,

    P. Lu, J. Lu, A. Liu, and G. Zhang, “Early concept drift detection via prediction uncertainty,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 18, 2025, pp. 19 124–19 132

    2025

  18. [18]

    Learning unbiased cluster descriptors for interpretable imbalanced concept drift detection,

    Y . Zhang, Z. Huang, M. Zhao, C. Zhang, Y . Lu, Y . Ji, F. Gu, and A. Zeng, “Learning unbiased cluster descriptors for interpretable imbalanced concept drift detection,”IEEE Transactions on Emerging Topics in Computational Intelligence, 2025

    2025

  19. [19]

    Just-in-time adaptive classifiers—part i: Detecting nonstationary changes,

    C. Alippi and M. Roveri, “Just-in-time adaptive classifiers—part i: Detecting nonstationary changes,”IEEE Transactions on Neural Net- works, vol. 19, no. 7, pp. 1145–1153, 2008

    2008

  20. [20]

    Unsupervised concept drift detection from deep learning representations in real-time,

    S. Greco, B. Vacchetti, D. Apiletti, and T. Cerquitelli, “Unsupervised concept drift detection from deep learning representations in real-time,” IEEE Transactions on Knowledge and Data Engineering, 2025

    2025

  21. [21]

    Kan4drift: Are kan effective for identifying and tracking concept drift in time series?

    K. Xu, L. Chen, and S. Wang, “Kan4drift: Are kan effective for identifying and tracking concept drift in time series?” inNeurIPS Workshop on Time Series in the Age of Large Models, 2024

    2024

  22. [22]

    Paired learners for concept drift,

    S. H. Bach and M. A. Maloof, “Paired learners for concept drift,” in 2008 Eighth IEEE International Conference on Data Mining. IEEE, 2008, pp. 23–32

    2008

  23. [23]

    Dynamic extreme learning machine for data stream classification,

    S. Xu and J. Wang, “Dynamic extreme learning machine for data stream classification,”Neurocomputing, vol. 238, pp. 433–449, 2017

    2017

  24. [24]

    Fp-elm: An online sequential learning algorithm for dealing with concept drift,

    D. Liu, Y . Wu, and H. Jiang, “Fp-elm: An online sequential learning algorithm for dealing with concept drift,”Neurocomputing, vol. 207, pp. 322–334, 2016

    2016

  25. [25]

    Adaptive random forests for evolving data stream classification,

    H. M. Gomes, A. Bifet, J. Read, J. P. Barddal, F. Enembreck, B. Pfharinger, G. Holmes, and T. Abdessalem, “Adaptive random forests for evolving data stream classification,”Machine Learning, vol. 106, no. 9, pp. 1469–1495, 2017

    2017

  26. [26]

    Online boosting adaptive learn- ing under concept drift for multistream classification,

    E. Yu, J. Lu, B. Zhang, and G. Zhang, “Online boosting adaptive learn- ing under concept drift for multistream classification,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 15, 2024, pp. 16 522–16 530

    2024

  27. [27]

    Coda: Temporal domain generalization via concept drift simulator,

    C.-Y . Chang, Y .-N. Chuang, Z. Jiang, K.-H. Lai, A. Jiang, and N. Zou, “Coda: Temporal domain generalization via concept drift simulator,” inProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 2, 2025, pp. 131–142

    2025

  28. [28]

    Training for the future: A simple gradient interpolation loss to generalize along time,

    A. Nasery, S. Thakur, V . Piratla, A. De, and S. Sarawagi, “Training for the future: A simple gradient interpolation loss to generalize along time,”Advances in Neural Information Processing Systems, vol. 34, pp. 19 198–19 209, 2021

    2021

  29. [30]

    Evolving standardization for continual domain generalization over temporal drift,

    M. Xie, S. Li, L. Yuan, C. Liu, and Z. Dai, “Evolving standardization for continual domain generalization over temporal drift,”Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  30. [31]

    Learning time-aware causal representation for model generalization in evolving domains,

    Z. He, S. Li, W. Song, L. Yuan, J. Liang, H. Li, and K. Gai, “Learning time-aware causal representation for model generalization in evolving domains,”arXiv preprint arXiv:2506.17718, 2025

    work page arXiv 2025

  31. [32]

    Autonomous online multistream generalization via fuzzy joint discriminant analysis,

    E. Yu, J. Lu, X. Yang, and G. Zhang, “Autonomous online multistream generalization via fuzzy joint discriminant analysis,”IEEE Transac- tions on Fuzzy Systems, vol. 34, no. 4, pp. 1256–1268, 2026

    2026

  32. [33]

    Learning to learn the future: Modeling concept drifts in time series prediction,

    X. You, M. Zhang, D. Ding, F. Feng, and Y . Huang, “Learning to learn the future: Modeling concept drifts in time series prediction,” in Proceedings of the 30th ACM International Conference on Information & Knowledge Management, 2021, pp. 2434–2443

    2021

  33. [34]

    Ddg-da: Data distribution generation for predictable concept drift adaptation,

    W. Li, X. Yang, W. Liu, Y . Xia, and J. Bian, “Ddg-da: Data distribution generation for predictable concept drift adaptation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 36, no. 4, 2022, pp. 4092–4100

    2022

  34. [36]

    Onenet: Enhancing time series forecasting models under concept drift by online ensembling,

    Q. Wen, W. Chen, L. Sun, Z. Zhang, L. Wang, R. Jin, T. Tanet al., “Onenet: Enhancing time series forecasting models under concept drift by online ensembling,”Advances in Neural Information Processing Systems, vol. 36, pp. 69 949–69 980, 2023. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 16

    2023

  35. [37]

    Proactive model adaptation against concept drift for online time series forecasting,

    L. Zhao and Y . Shen, “Proactive model adaptation against concept drift for online time series forecasting,” inProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 1, 2025, pp. 2020–2031

    2025

  36. [38]

    Delving into the Trajectory Long-tail Distribution for Muti-object Tracking,

    S. Chen, E. Yu, J. Li, and W. Tao, “Delving into the Trajectory Long-tail Distribution for Muti-object Tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 341–19 351

    2024

  37. [39]

    Long-Tailed Anomaly De- tection with Learnable Class Names,

    C.-H. Ho, K.-C. Peng, and N. Vasconcelos, “Long-Tailed Anomaly De- tection with Learnable Class Names,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12 435–12 446

    2024

  38. [40]

    The Neglected Tails in Vision-Language Models,

    S. Parashar, Z. Lin, T. Liu, X. Dong, Y . Li, D. Ramanan, J. Caverlee, and S. Kong, “The Neglected Tails in Vision-Language Models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12 988–12 997

    2024

  39. [41]

    Synthesizing Minority Samples for Long-tailed Classification via Distribution Matching,

    Z. Li, H. Zhao, J. Ren, A. Gao, D. Guo, X. Wan, and H. Zha, “Synthesizing Minority Samples for Long-tailed Classification via Distribution Matching,”Transactions on Machine Learning Research, Apr. 2025

    2025

  40. [42]

    Search and Detect: Training-Free Long Tail Object Detection via Web- Image Retrieval,

    M. Sidhu, H. Chopra, A. Blume, J. Kim, R. G. Reddy, and H. Ji, “Search and Detect: Training-Free Long Tail Object Detection via Web- Image Retrieval,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 15 129–15 138

    2025

  41. [43]

    Fewer tokens, greater scaling: Self-adaptive visual bases for efficient and expansive representation learning,

    S. Young, X. Zeng, and L. Xu, “Fewer tokens, greater scaling: Self-adaptive visual bases for efficient and expansive representation learning,”arXiv preprint arXiv:2511.19515, 2025

    work page arXiv 2025

  42. [44]

    DeiT-LT: Distillation Strikes Back for Vision Transformer Training on Long-Tailed Datasets,

    H. Rangwani, P. Mondal, M. Mishra, A. R. Asokan, and R. V . Babu, “DeiT-LT: Distillation Strikes Back for Vision Transformer Training on Long-Tailed Datasets,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 23 396–23 406

    2024

  43. [45]

    Long-Tail Class Incremental Learning via Independent Sub-prototype Construction,

    X. Wang, X. Yang, J. Yin, K. Wei, and C. Deng, “Long-Tail Class Incremental Learning via Independent Sub-prototype Construction,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 28 598–28 607

    2024

  44. [46]

    BEM: Balanced and Entropy-based Mix for Long-Tailed Semi-Supervised Learning,

    H. Zheng, L. Zhou, H. Li, J. Su, X. Wei, and X. Xu, “BEM: Balanced and Entropy-based Mix for Long-Tailed Semi-Supervised Learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 22 893–22 903

    2024

  45. [47]

    EIFA-KD: Explicit and implicit feature augmentation with knowledge distillation for long-tailed visual data classification,

    X. Deng, X. Wang, Y . Sun, X. Zhao, S. Tian, M. Li, and Y . Wang, “EIFA-KD: Explicit and implicit feature augmentation with knowledge distillation for long-tailed visual data classification,”Pattern Recogni- tion, vol. 171, p. 112129, Mar. 2026

    2026

  46. [48]

    T-distributed Spherical Feature Representation for Imbalanced Classification,

    X. Yang, Y . Chen, X. Yue, S. Xu, and C. Ma, “T-distributed Spherical Feature Representation for Imbalanced Classification,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 9, pp. 10 825–10 833, Jun. 2023

    2023

  47. [49]

    Adapting multi-modal large language model to concept drift from pre-training onwards,

    X. Yang, J. Lu, and E. Yu, “Adapting multi-modal large language model to concept drift from pre-training onwards,” inThe Thirteenth International Conference on Learning Representations, Y . Yue, A. Garg, N. Peng, F. Sha, and R. Yu, Eds., vol. 2025, 2025, pp. 90 869–90 891. [Online]. Available: https://proceedings.iclr.cc/paper files/paper/2025/ file/e25d8...

    2025

  48. [50]

    Domain Adaptation for Time Series Under Feature and Label Shifts,

    H. He, O. Queen, T. Koker, C. Cuevas, T. Tsiligkaridis, and M. Zitnik, “Domain Adaptation for Time Series Under Feature and Label Shifts,” inProceedings of the 40th International Conference on Machine Learning. PMLR, Jul. 2023, pp. 12 746–12 774

    2023

  49. [51]

    Gradual Domain Adaptation via Manifold-Constrained Distributionally Robust Optimization,

    A. Saberi, A. Najafi, A. Behjati, A. Emrani, Y . Zolfit, M. Shadrooy, A. Motahari, and B. H. Khalaj, “Gradual Domain Adaptation via Manifold-Constrained Distributionally Robust Optimization,” inAd- vances in Neural Information Processing Systems, vol. 37. Curran Associates, Inc., 2024, pp. 73 693–73 725

    2024

  50. [52]

    Open Compound Domain Adaptation with Object Style Compensation for Semantic Segmentation,

    T. Feng, H. Shi, X. Liu, W. Feng, L. Wan, Y . Zhou, and D. Lin, “Open Compound Domain Adaptation with Object Style Compensation for Semantic Segmentation,” inAdvances in Neural Information Process- ing Systems, vol. 36. Curran Associates, Inc., 2023, pp. 63 136–63 149

    2023

  51. [53]

    Subspace Identification for Multi-Source Domain Adaptation,

    Z. Li, R. Cai, G. Chen, B. Sun, Z. Hao, and K. Zhang, “Subspace Identification for Multi-Source Domain Adaptation,” inAdvances in Neural Information Processing Systems, vol. 36. Curran Associates, Inc., 2023, pp. 34 504–34 518

    2023

  52. [54]

    On f-Divergence Principled Domain Adaptation: An Improved Framework,

    Z. Wang and Y . Mao, “On f-Divergence Principled Domain Adaptation: An Improved Framework,” inAdvances in Neural Information Process- ing Systems, vol. 37. Curran Associates, Inc., 2024, pp. 6711–6748

    2024

  53. [55]

    Suffi- cient Invariant Learning for Distribution Shift,

    T. Kim, S. Park, S. Lim, Y . Jung, K. Muandet, and K. Song, “Suffi- cient Invariant Learning for Distribution Shift,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 4958–4967

    2025

  54. [56]

    Let Invariant Learning Inspire Neighbor-shift Generalization on Graphs,

    J. Li, J. Gao, B. Gu, and Y . Kong, “Let Invariant Learning Inspire Neighbor-shift Generalization on Graphs,”IEEE Transactions on Arti- ficial Intelligence, pp. 1–12, 2025

    2025

  55. [57]

    Graph Structure Extrapolation for Out- of-Distribution Generalization,

    X. Li, S. Gui, Y . Luo, and S. Ji, “Graph Structure Extrapolation for Out- of-Distribution Generalization,” inForty-First International Conference on Machine Learning, Jun. 2024

    2024

  56. [58]

    GeSS: Benchmarking Geometric Deep Learning under Scientific Applications with Distribution Shifts,

    D. Zou, S. Liu, S. Miao, V . Fung, S. Chang, and P. Li, “GeSS: Benchmarking Geometric Deep Learning under Scientific Applications with Distribution Shifts,”Advances in Neural Information Processing Systems, vol. 37, pp. 92 499–92 528, Dec. 2024

    2024

  57. [59]

    ReCDA: Concept Drift Adaptation with Representation Enhancement for Network Intrusion Detection,

    S. Yang, X. Zheng, J. Li, J. Xu, X. Wang, and E. C. H. Ngai, “ReCDA: Concept Drift Adaptation with Representation Enhancement for Network Intrusion Detection,” inProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, ser. KDD ’24. New York, NY , USA: Association for Computing Machinery, Aug. 2024, pp. 3818–3828

    2024

  58. [60]

    Test- Time Adaptation Induces Stronger Accuracy and Agreement-on-the- Line,

    E. Kim, M. Sun, C. Baek, A. Raghunathan, and J. Z. Kolter, “Test- Time Adaptation Induces Stronger Accuracy and Agreement-on-the- Line,”Advances in Neural Information Processing Systems, vol. 37, pp. 120 184–120 220, Dec. 2024

    2024

  59. [61]

    Dual Test-Time Training for Out-of-Distribution Recom- mender System,

    X. Yang, Y . Wang, J. Chen, W. Fan, X. Zhao, E. Zhu, X. Liu, and D. Lian, “Dual Test-Time Training for Out-of-Distribution Recom- mender System,”IEEE Transactions on Knowledge and Data Engi- neering, vol. 37, no. 6, pp. 3312–3326, Jun. 2025

    2025

  60. [62]

    OODD: Test-time Out-of-Distribution Detection with Dynamic Dictionary,

    Y . Yang, L. Zhu, Z. Sun, H. Liu, Q. Gu, and N. Ye, “OODD: Test-time Out-of-Distribution Detection with Dynamic Dictionary,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 30 630–30 639

    2025

  61. [63]

    UDDA-TC: Unsupervised Real-Time Drift Detection and Adaptation for Continual Traffic Classification in Mobile Edge Computing,

    M. Liu, P. Wang, Z. Li, Y . Ye, Z. Wang, and X. Chen, “UDDA-TC: Unsupervised Real-Time Drift Detection and Adaptation for Continual Traffic Classification in Mobile Edge Computing,”IEEE Transactions on Consumer Electronics, pp. 1–1, 2025

    2025

  62. [64]

    AdaO2B: Adaptive Online to Batch Conversion for Out-of-Distribution Generalization,

    X. Zhang, S. Dai, J. Xu, Y . Liu, and Z. Dong, “AdaO2B: Adaptive Online to Batch Conversion for Out-of-Distribution Generalization,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 21, pp. 22 596–22 604, Apr. 2025

    2025

  63. [65]

    Label-Noise Robust Diffusion Models,

    B. Na, Y . Kim, H. Bae, J. H. Lee, S. J. Kwon, W. Kang, and I.-c. Moon, “Label-Noise Robust Diffusion Models,” inThe Twelfth International Conference on Learning Representations, Oct. 2023

    2023

  64. [66]

    Optimizing OOD Detection in Molecular Graphs: A Novel Approach with Diffusion Models,

    X. Shen, Y . Wang, K. Zhou, S. Pan, and X. Wang, “Optimizing OOD Detection in Molecular Graphs: A Novel Approach with Diffusion Models,” inProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, ser. KDD ’24. New York, NY , USA: Association for Computing Machinery, Aug. 2024, pp. 2640– 2650

    2024

  65. [67]

    Benchmark- ing common uncertainty estimation methods with histopathological images under domain shift and label noise,

    H. A. Mehrtens, A. Kurz, T.-C. Bucher, and T. J. Brinker, “Benchmark- ing common uncertainty estimation methods with histopathological images under domain shift and label noise,”Medical Image Analysis, vol. 89, p. 102914, Oct. 2023

    2023

  66. [68]

    MaxEnt Loss: Constrained Max- imum Entropy for Calibration under Out-of-Distribution Shift,

    D. Neo, S. Winkler, and T. Chen, “MaxEnt Loss: Constrained Max- imum Entropy for Calibration under Out-of-Distribution Shift,”Pro- ceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 19, pp. 21 463–21 472, Mar. 2024

    2024

  67. [69]

    Modeling the Distributional Uncertainty for Salient Object Detection Models,

    X. Tian, J. Zhang, M. Xiang, and Y . Dai, “Modeling the Distributional Uncertainty for Salient Object Detection Models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 19 660–19 670

    2023

  68. [70]

    FedStream: Prototype-Based Federated Learning on Distributed Concept-Drifting Data Streams,

    C. B. Mawuli, L. Che, J. Kumar, S. U. Din, Z. Qin, Q. Yang, and J. Shao, “FedStream: Prototype-Based Federated Learning on Distributed Concept-Drifting Data Streams,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 53, no. 11, pp. 7112– 7124, Nov. 2023

    2023

  69. [71]

    FOOGD: Federated Collaboration for Both Out-of- distribution Generalization and Detection,

    X. Liao, W. Liu, P. Zhou, F. Yu, J. Xu, J. Wang, W. Wang, C. Chen, and X. Zheng, “FOOGD: Federated Collaboration for Both Out-of- distribution Generalization and Detection,”Advances in Neural Infor- mation Processing Systems, vol. 37, pp. 132 908–132 945, Dec. 2024

    2024

  70. [72]

    FedGOG: Federated Graph Out-of- Distribution Generalization with Diffusion Data Exploration and Latent Embedding Decorrelation,

    P. Zhou, C. Chen, W. Liu, X. Liao, W. Shen, J. Xu, Z. Fu, J. Wang, W. Wen, and X. Zheng, “FedGOG: Federated Graph Out-of- Distribution Generalization with Diffusion Data Exploration and Latent Embedding Decorrelation,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 21, pp. 22 965–22 973, Apr. 2025

    2025

  71. [73]

    FedFM: A federated few-shot learning method by comparison network and model calibration,

    C. Zhao, S. Bao, M. Chen, Z. Gao, K. Xiao, and P. Dai, “FedFM: A federated few-shot learning method by comparison network and model calibration,”Knowledge-Based Systems, vol. 309, p. 112848, Jan. 2025. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 17

    2025

  72. [74]

    Concept drift detection with quadtree-based spatial mapping of streaming data,

    R. Amador Coelho, L. C. Bambirra Torres, and C. Leite de Cas- tro, “Concept drift detection with quadtree-based spatial mapping of streaming data,”Information Sciences, vol. 625, pp. 578–592, May 2023

    2023

  73. [75]

    Unsupervised Unlearning of Concept Drift with Autoen- coders,

    A. Artelt, K. Malialis, C. G. Panayiotou, M. M. Polycarpou, and B. Hammer, “Unsupervised Unlearning of Concept Drift with Autoen- coders,” in2023 IEEE Symposium Series on Computational Intelli- gence (SSCI), Dec. 2023, pp. 703–710

    2023

  74. [76]

    Class Incre- mental Learning based on Identically Distributed Parallel One-Class Classifiers,

    W. Sun, Q. Li, J. Zhang, W. Wang, and Y .-a. Geng, “Class Incre- mental Learning based on Identically Distributed Parallel One-Class Classifiers,”Neurocomputing, vol. 556, p. 126579, Nov. 2023

    2023

  75. [77]

    FeTrIL: Feature Translation for Exemplar-Free Class-Incremental Learning,

    G. Petit, A. Popescu, H. Schindler, D. Picard, and B. Delezoide, “FeTrIL: Feature Translation for Exemplar-Free Class-Incremental Learning,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 3911–3920

    2023

  76. [78]

    Dynamic Integration of Task-Specific Adapters for Class Incremental Learning,

    J. Li, S. Wang, B. Qian, Y . He, X. Wei, Q. Wang, and Y . Gong, “Dynamic Integration of Task-Specific Adapters for Class Incremental Learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 30 545–30 555

    2025

  77. [79]

    Antiforgetting Incremental Learning Algorithm for Interval Type-2 Fuzzy Neural Network,

    C. Sun, H. Han, X. Wu, and H. Yang, “Antiforgetting Incremental Learning Algorithm for Interval Type-2 Fuzzy Neural Network,”IEEE Transactions on Fuzzy Systems, vol. 32, no. 4, pp. 1938–1950, Apr. 2024

    1938

  78. [80]

    Concept Drift Adaptation by Exploiting Drift Type,

    J. Li, H. Yu, Z. Zhang, X. Luo, and S. Xie, “Concept Drift Adaptation by Exploiting Drift Type,”ACM Trans. Knowl. Discov. Data, vol. 18, no. 4, pp. 96:1–96:22, Feb. 2024

    2024

  79. [81]

    MalFSCIL: A Few-Shot Class-Incremental Learning Approach for Malware Detection,

    Y . Chai, X. Chen, J. Qiu, L. Du, Y . Xiao, Q. Feng, S. Ji, and Z. Tian, “MalFSCIL: A Few-Shot Class-Incremental Learning Approach for Malware Detection,”IEEE Transactions on Information Forensics and Security, vol. 20, pp. 2999–3014, 2025

    2025

  80. [82]

    Generalized incremental learning under concept drift across evolving data streams,

    E. Yu, J. Lu, and G. Zhang, “Generalized incremental learning under concept drift across evolving data streams,”Proceedings of the ACM Web Conference 2026, p. 3905–3916, 2026

    2026

Showing first 80 references.