Pith. sign in

REVIEW 3 major objections 7 minor 41 references

5G-DIL: Domain Incremental Learning with Similarity-Aware Sampling for Dynamic 5G Indoor Localization

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A similarity-aware sampling method adapts 5G indoor positioning to changed environments with as few as 50 exemplars, the paper claims.

desk verdict Useful empirical core—similarity-based exemplar selection beats random/equal replay on real 5G CIR data—but the abstract oversells it: 0.261 m is N=200, not 50, and the 'modified-regions-only' pipeline is never actually tested. read the letter →

arxiv 2505.17684 v1 pith:YR7DOQJW submitted 2025-05-23 cs.CV

Nisha Lakshmana Raichur , Lucas Heublein , Christopher Mutschler , Felix Ott This is my paper · ORCID
classification cs.CV
keywords 5Gindoorlocalizationdomainincrementallearningexemplarselectionsimilarity-awaresamplingChebyshevdistancecatastrophicforgettingchannelimpulseresponsefingerprintingtransfer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that a domain incremental learning method, 5G-DIL, can adapt a 5G fingerprint positioning model to a changed indoor environment without retraining on the whole new site. It selects a small set of exemplars from the new domain that most resemble the old domain using distance-based similarity, and fine-tunes only on modified regions while averaging weights to avoid catastrophic forgetting. The authors report a mean positioning error of 0.261 meters with as few as 50 exemplars, which is close to the accuracy of full retraining while reducing the amount of adaptation data by orders of magnitude.

What carries the argument

The load-bearing mechanism is similarity-aware exemplar sampling: for every candidate sample in the adaptation domain, the method finds its nearest neighbor in the previous domain under a distance metric, selects the most similar samples as exemplars, and then trains only on the modified regions of the new environment while averaging the current and previous model weights. The Chebyshev distance emerges as the best-performing metric, and the KDTree nearest-neighbor search makes the similarity computation fast enough for practical use.

What would settle it

Run the same 5G-DIL pipeline on a dataset where changed and unchanged regions are explicitly annotated, then compare adaptation when exemplars are drawn only from the changed regions or when the region labels are wrong; if the error rises toward the full-retrain baseline or forgetting appears, the changed-region assumption is the cause.

Watch

Extended reading notes

Core claim

The central claim is that similarity-aware exemplar selection identifies the most representative samples from a new environment so that a domain incremental learning model needs no more than 200, and sometimes as few as 50, exemplars to keep the localization error low. The paper shows that selecting the top-N most similar samples between the previous and adaptation domains, measured with the Chebyshev distance and computed efficiently with a KDTree, outperforms random, equally distributed, and error-dependent selection. On their real-world 5G dataset, the best configuration achieves a mean absolute error of 0.261 meters under dynamic environmental conditions while preserving accuracy on previous environments, and the method works across both deterministic and realistic non-line-of-sight sites.

Load-bearing premise

The method assumes someone already knows which regions of the new environment are modified and which are static, because the model is trained only on the modified regions while exemplars are picked from the static ones.

Editorial extensions

If this is right

  • Adapting a 5G fingerprinting model to a new environment would require only tens to hundreds of labeled samples instead of the tens of thousands typically used, cutting data collection cost sharply.
  • Learning-without-forgetting and elastic weight consolidation benefit most from similarity-based exemplars, reaching lower errors than random or error-dependent sample selection.
  • The choice of distance metric matters: cosine similarity performs poorly, while Chebyshev, Manhattan, and Canberra distances all give strong results.
  • A full fine-tune takes about 500 seconds per adaptation in their setup, whereas the exemplar-based LwF and EWC variants take about 85 to 100 seconds, so the approach also reduces compute time.
  • Because the method is evaluated across five different indoor sites with LoS and NLoS conditions, the authors argue it extends to practical industrial deployments where moving objects change the radio environment.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One consequence the authors leave implicit is that the practical bottleneck shifts from data labeling to change detection: the 50-exemplar result depends on knowing which regions are modified, so a deployment would need a reliable way to identify those regions first.
  • A testable extension is to measure how the required exemplar count grows as the fraction of changed area increases; the current experiments treat whole sites as domains, so they do not directly quantify this relationship.
  • The same exemplar-selection principle could plausibly transfer to other fingerprint signals such as WiFi or UWB, where environmental dynamics also force costly retraining, but the distance-metric ranking would need to be revalidated for each signal type.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper proposes 5G-DIL, a domain incremental learning approach for 5G indoor localization that adapts a model to new environments by fine-tuning on a small set of exemplars selected through a similarity-aware sampling procedure. The method computes distances between channel impulse responses in the previous and adaptation domains using a KDTree, selects the top-N most similar samples from the adaptation domain, and combines these exemplars with regularized fine-tuning (EWC, LwF, SI), weight averaging, and comparisons against Finetune and PNN. Experiments on a real-world five-site dataset (RS3, RS4, RS5, DS1, DS2) report MAE positioning errors for various exemplar counts (N = 0, 50, 100, 200) and several sampling strategies. The headline claims are that the method 'requires as few as 50 exemplars' and 'achieves an MAE positioning error of 0.261 meters', and that it 'trains only on the modified regions of the new environment.'

Significance. If fully supported, the method would be practically significant: it promises to cut the labeled data needed for adapting 5G fingerprinting models from tens of thousands of samples to a few hundred, while maintaining near-baseline accuracy. The paper's strengths include a real-world evaluation with multiple DIL baselines and sampling strategies, reproducible experimental setup, and a code release. The experiments do support the weaker claim that similarity-aware sampling improves over random and equal-distribution selection for a fixed exemplar budget. However, the headline quantitative claim and the 'modified-regions-only' training claim are not supported by the reported experiments, which limits the paper's contribution as currently stated.

major comments (3)
  1. [Abstract and §VI (Conclusion)] The abstract states that the method 'requires as few as 50 exemplars' and 'achieves an MAE positioning error of 0.261 meters'; the conclusion repeats 'errors of 0.261 m for small changes and 0.266 m for larger changes.' These numbers do not correspond to N = 50. In Table II, the 0.261 m value appears in the RS5 row, N = 200 sub-row, LwF column, and the 0.266 m value appears in the DS1 row, N = 100 sub-row, LwF column. At N = 50, the best MAE across methods is about 0.385 m (RS4, EWC), with larger values on RS5, DS1, and DS2. The 'as few as 50' claim is therefore not supported by the reported results. The paper should either report the accuracy achievable at N = 50 or clearly associate the headline accuracy with the larger exemplar counts.
  2. [§III (Method Overview) and Eq. (2) with Table I] The central 'modified-regions-only' efficiency claim is neither implemented nor tested. Section III states that 'the model is trained only on the modified regions of the new scenario' while exemplars are selected from 'non-modified regions.' However, Eq. (2) computes distances between samples of the previous domain D_{t-1} and all samples of the adaptation domain D_t, and the selection takes the top-N samples from D_t with highest similarity to D_{t-1}, i.e., the least-changed samples. If these selected samples are the only adaptation training data (as Table I's 'adaptation (subset with N exemplars)' caption implies), then the model is trained on the least-changed samples, not on the modified regions. If, instead, the model is trained on all modified regions plus these exemplars, no experiment provides a partition of any site into modified and unchanged regions; each site (RS3, RS4, RS5, DS1, DS2) is treated as a whole domain. Section VI's statement that the framework 'collects data only from dynamic regions' is therefore not evidenced. The authors should add experiments with region-level annotations, or re-scope the claims to whole-domain adaptation with similarity-based exemplar selection and explicitly state the assumption that change detection infrastructure exists.
  3. [§V (Evaluation of Sample Selection) and Fig. 3] The distance metric for similarity-aware selection is chosen after inspecting results on the same evaluation domains: Figure 3 compares eight distance metrics on the adaptation RS3 → RS4 → RS5 and reports errors on RS4/RS5, and the 'best' metric is then used for the final results in Table II. Unless a validation split was used to select the metric (which is not described), this constitutes selection on the test set and may overstate the advantage of Chebyshev/Manhattan over random or ED sampling. Please clarify whether the metric was chosen on held-out validation data or provide a cross-validation procedure.
minor comments (7)
  1. [§III (Notation)] The term 'exemplars' is used for samples selected from the adaptation domain that are most similar to the previous domain, whereas in standard continual learning 'exemplars' are stored samples from previous tasks. Please clarify the terminology to avoid confusion.
  2. [§V (DIL Evaluation)] The text says 'For N ∈ {50, 100}, LwF achieves the lowest positioning error,' but Table II shows that at N = 50, EWC outperforms LwF on RS4 (0.385 m vs. 0.493 m) and is competitive on other domains. Please qualify this statement with the actual per-domain comparisons.
  3. [§VI (Conclusion)] The conclusion's distinction between 'small changes' and 'larger changes' is not defined. Which sites or adaptation steps correspond to these categories? Without a definition, the reader cannot interpret the 0.261 m and 0.266 m values.
  4. [Eq. (2)] Equation (2) does not specify whether the CIRs are preprocessed (e.g., normalized, truncated, or aligned) before computing distances. Please describe the input representation used in the distance computation.
  5. [Fig. 1 and §III (Method Overview)] The figure mentions 'weight averaging' but the text does not provide an equation or a precise description of how weights are averaged after each task. Please specify the averaging scheme (e.g., arithmetic mean with or without task-specific scaling).
  6. [Table II] Table II is dense and difficult to parse because each row combines five domains and five methods. Consider splitting the results by method or using a different layout, and report the standard deviation (the text says mean and standard variance are computed).
  7. [§V (Computation Times)] The sentence 'PNN expands the network with each new task (250 s)' lists a time under 300 s, but the text says FT requires 500 s; it may be useful to clarify whether the reported PNN time includes the lateral connections or only the new task's training.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the reported results are empirical benchmark outcomes, not derivations that reduce to their inputs; the only concerns are an untested region-prior assumption and test-set-informed metric selection, neither of which is circular by construction.

full rationale

5G-DIL is an empirical study rather than a derivation chain. The similarity-aware selection in Eq. (2) computes distances between previous-domain CIR samples and candidate adaptation-domain samples, and the top-N most similar exemplars are used for fine-tuning; the reported MAE values (e.g., 0.261 m in Table II: RS5, LwF, N=200) are measured on the test domains, not derived from the selection rule. The paper does not fit a parameter and then rename that fit as a prediction, nor does it import a uniqueness result from the authors' prior work. The self-citations ([13], [23], [30], [35], [36]) are background and not load-bearing. The central efficiency claim ('training only on the modified regions') is not actually implemented or tested: Eq. (2) selects the least-changed samples, and each site (RS3, RS4, RS5, DS1, DS2) is treated as a whole domain, so the modified-regions-only pipeline is a stated assumption rather than a tested mechanism. This is an unresolved validation gap, not a circular reduction. Similarly, selecting the Chebyshev/Manhattan distance after inspecting Figure 3 on the same adaptation domains is benchmark selection, not a fitted input renamed as a prediction. No circular step can be exhibited with the required specificity; the core empirical comparison is self-contained against the Stahlke et al. dataset and external DIL baselines.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper's empirical claims rest on several tuned hyperparameters and on an untested assumption about known region changes. There are no invented physical entities.

free parameters (5)
  • Regularization strength lambda for EWC = 100,000 (best for small exemplar sets)
    Tuned in Fig. 5a on RS3 test data; controls penalty on important weights.
  • Regularization strength lambda for LwF = 10.0 (best for small exemplar sets)
    Tuned in Fig. 5b on RS3 test data.
  • Regularization strength lambda for SI = 5.0 to 10.0
    Tuned in Fig. 5c; higher lambda is better for exemplar sets larger than 50.
  • Exemplar set size N = 0, 50, 100, 200
    Chosen by hand; the abstract emphasizes 50 but best results often use 200.
  • Distance metric = Chebyshev for LwF, Manhattan in trajectory figure, ED-MSEmin for EWC
    Selected post hoc by comparing eight metrics on the same evaluation domains, introducing selection bias.
assumptions (4)
  • domain assumption Changed regions in a new environment are known or identifiable in advance
    Introduction states 'an existing infrastructure for detecting and localizing changes is often already in place'; the method trains only on modified regions and uses exemplars from unchanged ones.
  • domain assumption CIR statistics of unchanged regions are stationary across domains
    Similarity-based exemplars from D_{t-1} are assumed to remain representative in D_t; otherwise the selected exemplars would not preserve prior knowledge.
  • ad hoc to paper Nearest neighbors in CIR feature space are the most useful exemplars for position regression
    The core sampling rule, Eq. 2, equates similarity in raw CIR space with training value, an assumption validated only empirically on the five sites.
  • standard math MSE on 2D position is the correct objective for continual localization
    Used as loss and evaluation metric; standard but not discussed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 5G-DIL: Domain Incremental Learning with Similarity-Aware Sampling for Dynamic 5G Indoor Localization." pith.science (2026). https://pith.science/paper/YR7DOQJW

@misc{pith2026250517684,
  author       = {Pith},
  title        = {Pith review of: 5G-DIL: Domain Incremental Learning with Similarity-Aware Sampling for Dynamic 5G Indoor Localization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YR7DOQJW}},
  note         = {Machine review of arXiv:2505.17684}
}
read the original abstract

Indoor positioning based on 5G data has achieved high accuracy through the adoption of recent machine learning (ML) techniques. However, the performance of learning-based methods degrades significantly when environmental conditions change, thereby hindering their applicability to new scenarios. Acquiring new training data for each environmental change and fine-tuning ML models is both time-consuming and resource-intensive. This paper introduces a domain incremental learning (DIL) approach for dynamic 5G indoor localization, called 5G-DIL, enabling rapid adaptation to environmental changes. We present a novel similarity-aware sampling technique based on the Chebyshev distance, designed to efficiently select specific exemplars from the previous environment while training only on the modified regions of the new environment. This avoids the need to train on the entire region, significantly reducing the time and resources required for adaptation without compromising localization accuracy. This approach requires as few as 50 exemplars from adaptation domains, significantly reducing training time while maintaining high positioning accuracy in previous environments. Comparative evaluations against state-of-the-art DIL techniques on a challenging real-world indoor dataset demonstrate the effectiveness of the proposed sample selection method. Our approach is adaptable to real-world non-line-of-sight propagation scenarios and achieves an MAE positioning error of 0.261 meters, even under dynamic environmental conditions. Code: https://gitlab.cc-asp.fraunhofer.de/5g-pos/5g-dil

Figures

Figures reproduced from arXiv: 2505.17684 by the authors.

Figure 1
Figure 1. Overview of our 5G-DIL method. The model is trained using CIR data Dt and the corresponding position labels P t for each task t within the 5G dataset [2]. To incrementally learn across domains without incurring catastrophic forgetting, the model updates its parameters by averaging the weights obtained from previous tasks {t−1, t−2, . . . , 1}. Our similarity-aware sample selection strategy is employed to choose a li… view at source ↗
Figure 2
Figure 2. Sample selection for task Dt (red: environmental changes, blue: static environment, yellow dots: selected samples). from the adaptation domain of task t that optimally represent the domain characteristics and capture domain shifts, thereby minimizing positioning error while ensuring efficient fine￾tuning runtime. The baseline approach involves randomly distributed (RD) selecting N samples from the domain dataset bas… view at source ↗
Figure 3
Figure 3. Evaluation of distance metrics for similarity-aware [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Predicted trajectories and the corresponding errors in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Hyperparameter search of λ for the adaptation RS3 → RS4 on the RS3 test dataset for EWC, LwF, and SI. controlling the balance between retaining knowledge from previous tasks and acquiring new ones, penalizing significant deviations of important weights. A higher value …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 36 canonical work pages

  1. [1]

    A Comprehensive Framework for 5G Indoor Localization,

    A. L. Floch, R. Kacimi, P. Druart, Y . Lefebvre, and Andr ´e, “A Comprehensive Framework for 5G Indoor Localization,” in Computer Communications, Dec. 2024

    work page 2024

  2. [2]

    Transfer Learning to Adapt 5G AI-based Fingerprint Localization Across Environments,

    M. Stahlke, T. Feigl, M. H. C. Garc ´ıa, R. A. Stirling-Gallacher, J. Seitz, and C. Mutschler, “Transfer Learning to Adapt 5G AI-based Fingerprint Localization Across Environments,” in IEEE V ehicular Technology Conf. (VTC2022-Spring), Helsinki, Finland, Jun. 2022

    work page 2022

  3. [3]

    Wiometrics: Comparative Performance of Artificial Neural Networks for Wireless Navigation,

    R. Whiton, J. Chen, and F. Tufvesson, “Wiometrics: Comparative Performance of Artificial Neural Networks for Wireless Navigation,” in IEEE Trans. on V ehicular Technology (TVT) , vol. 73(9), Sep. 2024

    work page 2024

  4. [4]

    5G Positioning Advancements with AI/ML,

    M. Alawieh and G. Kontes, “5G Positioning Advancements with AI/ML,” in arXiv preprint arXiv:2401.02427 , Nov. 2023

    arXiv 2023

  5. [5]

    Multi-Environment based Meta-Learning with CSI Fingerprints for Radio Based Positioning,

    A. Foliadis, M. H. C. Garc ´ıa, R. A. Stirling-Gallacher, and R. S. Thom¨a, “Multi-Environment based Meta-Learning with CSI Fingerprints for Radio Based Positioning,” in IEEE Wireless Communications and Networking Conf. (WCNC) , Glasgow, UK, Mar. 2023

    work page 2023

  6. [6]

    FeMLoc: Federated Meta- learning for Adaptive Wireless Indoor Localization Tasks in IoT Net- works,

    Y . Etiabi, W. Njima, and E. M. Amhoud, “FeMLoc: Federated Meta- learning for Adaptive Wireless Indoor Localization Tasks in IoT Net- works,” in IEEE Internet of Things Journal , vol. 11(22), Aug. 2024

    work page 2024

  7. [7]

    Mitigating Catastrophic Forgetting in Deep Transfer Learning for Fingerprinting Indoor Posi- tioning,

    H. Pan, S. Wei, D. He, Z. Xiao, and S. Arai, “Mitigating Catastrophic Forgetting in Deep Transfer Learning for Fingerprinting Indoor Posi- tioning,” in IEEE Intl. Symposium on Circuits and Systems (ISCAS) , Monterey, CA, May 2023

    work page 2023

  8. [8]

    Uncovering the Potential of Indoor Localization: Role of Deep and Transfer Learning,

    O. Kerdjidj, Y . Himeur, S. S. Sohail, A. Amira, F. Fadli, and S. Atalla, “Uncovering the Potential of Indoor Localization: Role of Deep and Transfer Learning,” in IEEE Access , vol. 12, May 2024, pp. 73 980– 74 010

    work page 2024

Show all 41 references
  1. [9]

    Indoor Localization in Commercial 5G Environment with Single BS,

    Y . Dai, L. Chen, X. Zhou, Y . Ruan, and R. Chen, “Indoor Localization in Commercial 5G Environment with Single BS,” in Int. Arch. Pho- togramm. Remote Sens. Spatial Inf. Sci. , 2024

    2024

  2. [10]

    Deep Learning-based Positioning with Multi-task Learning and Uncertainty-based Fusion,

    A. Foliadis, M. H. C. Garc ´ıa, R. A. Stirling-Gallacher, and R. S. Thom¨a, “Deep Learning-based Positioning with Multi-task Learning and Uncertainty-based Fusion,” in IEEE Trans. on Machine Learning in Communications and Networking (TMLCN) , vol. 2, Aug. 2024, pp. 1127–1141

    2024

  3. [11]

    A Tutorial on Terahertz-Band Localization for 6G Communication Systems,

    H. Chen, H. Sarieddeen, T. Ballal, H. Wymeersch, M.-S. Alouini, and T. Y . Al-Naffouri, “A Tutorial on Terahertz-Band Localization for 6G Communication Systems,” in IEEE Communications Surveys & Tutorials (COMST), vol. 24(3), May 2022, pp. 1780–1815

    2022

  4. [12]

    A CSI- Based Data-Driven Localization Framework Using Small-Scale Training Datasets in Single-Site MIMO Systems,

    K. Vuckovic, S. Hosseini, F. Hejazi, and N. Rahnavard, “A CSI- Based Data-Driven Localization Framework Using Small-Scale Training Datasets in Single-Site MIMO Systems,” in IEEE Trans. on Wireless Communications (TWC) , vol. 23(11), Aug. 2024, pp. 16 346–16 358

    2024

  5. [13]

    Evaluating ML Robustness in GNSS Interference Classification, Characterization & Localization,

    L. Heublein, T. Feigl, T. Nowak, A. R ¨ugamer, C. Mutschler, and F. Ott, “Evaluating ML Robustness in GNSS Interference Classification, Characterization & Localization,” in arXiv preprint arXiv:2409.15114 , Apr. 2025

    2025 arXiv

  6. [14]

    5G1M: Indoor Fingerprint Positioning Using a Single 5G Module,

    C. Wang, X. Jin, C. Xia, C. Xu, Y . Sun, and Y . Duan, “5G1M: Indoor Fingerprint Positioning Using a Single 5G Module,” in IEEE Sensors Journal, Sep. 2024

    2024

  7. [15]

    RSSI-based Fingerprint Localization in LoRaW AN Networks Using CNNs with Squeeze and Excitation Blocks,

    A. S. Lutakamale, H. C. Myburgh, and A. de Freitas, “RSSI-based Fingerprint Localization in LoRaW AN Networks Using CNNs with Squeeze and Excitation Blocks,” in Ad Hoc Networks , vol. 159, Jun. 2024

    2024

  8. [16]

    SNWPM: A Siamese Network Based Wireless Positioning Model Resilient to Partial Base Stations Unavailable,

    Y . Zhu, J. Wang, Y . Sun, B. Xu, P. Liu, and Z. Pan, “SNWPM: A Siamese Network Based Wireless Positioning Model Resilient to Partial Base Stations Unavailable,” in China Communications , vol. 20(9), Sep. 2023, pp. 20–33

    2023

  9. [17]

    Im- plementation of a Transfer Learning and Fingerprint-Based Positioning System,

    N.-I. Kim, T.-J. Kim, Y . Moon, Y . H. Jung, J.-H. Na, and H. Kim, “Im- plementation of a Transfer Learning and Fingerprint-Based Positioning System,” in IEEE V ehicular Technology Conf. (VTC-Fall), Washington, DC, Oct. 2024

    2024

  10. [18]

    An Attention Auxiliary Network- Based Method for WiFi Fingerprint Indoor Localization,

    Q. Fu, W. Liu, Y . Chen, and Z. Deng, “An Attention Auxiliary Network- Based Method for WiFi Fingerprint Indoor Localization,” in Intl. Conf. on Intelligent Computing and Signal Processing (ICSP) , Xian, China, Apr. 2024

    2024

  11. [19]

    Progressive Neural Networks,

    A. A. Rusu, N. C. Rabinowitz, G. Desjardins, H. Soyer, J. Kirkpatrick, K. Kavukcuoglu, R. Pascanu, and R. Hadsell, “Progressive Neural Networks,” in arXiv preprint arXiv:1606.04671 , Oct. 2022

    2022 arXiv

  12. [20]

    Continual Learning Through Synaptic Intelligence,

    F. Zenke, B. Poole, and S. Ganguli, “Continual Learning Through Synaptic Intelligence,” in Intl. Conf. on Machine Learning (ICML) , vol. 70, Aug. 2017, pp. 3987–3995

    2017

  13. [21]

    Experi- ence Replay for Continual Learning,

    D. Rolnick, A. Ahuja, J. Schwarz, T. Lillicrap, and G. Wayne, “Experi- ence Replay for Continual Learning,” in Advances in Neural Information Processing Systems (NIPS) , 2019

    2019

  14. [22]

    Overcoming Catastrophic Forgetting in Neural Networks,

    J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Desjardins, A. A. Rusu, K. Milan, J. Quan, T. Ramalho, A. Grabska-Barwinska, D. Hassabis, C. Clopath, D. Kumaran, and R. Hadsell, “Overcoming Catastrophic Forgetting in Neural Networks,” in Applied Mathematics , vol. 114...

    2017

  15. [23]

    Federated Learning with MMD-based Early Stopping for Adaptive GNSS Interference Classification,

    N. S. Gaikwad, L. Heublein, N. L. Raichur, T. Feigl, C. Mutschler, and F. Ott, “Federated Learning with MMD-based Early Stopping for Adaptive GNSS Interference Classification,” in arXiv preprint arXiv:2410.15681v2, Dec. 2024

    2024 arXiv

  16. [24]

    Learning Without Forgetting,

    Z. Li and D. Hoiem, “Learning Without Forgetting,” in IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI) , vol. 40(12), Dec. 2018, pp. 2935–2947

    2018

  17. [25]

    DTL-5G: Deep Transfer Learning-based DDoS Attack Detection in 5G and Beyond Networks,

    B. Farzaneh, N. Shahriar, A. H. A. Muktadir, M. S. Towhid, and M. S. Khosravani, “DTL-5G: Deep Transfer Learning-based DDoS Attack Detection in 5G and Beyond Networks,” in Computer Communications, vol. 228, Dec. 2024

    2024

  18. [26]

    Dynamic Anomaly Detection in 5G-Connected IoT Devices using Transfer Learning,

    M. Dwedar, F. Bayram, J. Eberhard, and A. Jesser, “Dynamic Anomaly Detection in 5G-Connected IoT Devices using Transfer Learning,” in IEEE Intl. Conf. on Computing and Machine Intelligence (ICMI) , Mt Pleasant, MI, Apr. 2024

    2024

  19. [27]

    Improving Replay Sample Selection and Storage for Less Forgetting in Continual Learning,

    D. Brignac, N. Lobo, and A. Mahalanobis, “Improving Replay Sample Selection and Storage for Less Forgetting in Continual Learning,” in IEEE Intl. Conf. on Computer Vision Workshops (ICCVW) , 2024, pp. 3540–3549

    2024

  20. [28]

    One-Shot Domain Incremental Learning,

    Y . Esaki, S. Koide, and T. Kutsuna, “One-Shot Domain Incremental Learning,” in Advances in Neural Information Processing Systems (NIPS) Workshops, Mar. 2024

    2024

  21. [29]

    An Efficient Domain-Incremental Learning Approach to Drive in All Weather Condi- tions,

    M. J. Mirza, M. Masana, H. Possegger, and H. Bischof, “An Efficient Domain-Incremental Learning Approach to Drive in All Weather Condi- tions,” in IEEE Intl. Conf. on Computer Vision and Pattern Recognition Workshops (CVPRW), New Orleands, LA, Jun. 2022

    2022

  22. [30]

    Bayesian Learning-driven Prototypical Contrastive Loss for Class-Incremental Learning,

    N. L. Raichur, L. Heublein, T. Feigl, A. R ¨ugamer, C. Mutschler, and F. Ott, “Bayesian Learning-driven Prototypical Contrastive Loss for Class-Incremental Learning,” in Trans. on Machine Learning Research (TMLR), Apr. 2025

    2025

  23. [31]

    Graph-Based vs. Error State Kalman Filter-Based Fusion of 5G and Inertial Data for MA V Indoor Pose Estimation,

    M. Kabiri, C. Cimarelli, H. Bavle, J. L. Sanchez-Lopez, and H. V ooos, “Graph-Based vs. Error State Kalman Filter-Based Fusion of 5G and Inertial Data for MA V Indoor Pose Estimation,” inJournal of Intelligent & Robotic Systems , vol. 110(87), Jun. 2024

    2024

  24. [32]

    Neural 5G Indoor Localization with IMU Supervision,

    A. Ermolov, S. Kadambi, M. Arnold, M. Hirzallah, R. Amiri, and D. S. M. Singh, “Neural 5G Indoor Localization with IMU Supervision,” in IEEE Global Communications Conf. (GLOBECOM) , Kuala Lumpur, Malaysia, Dec. 2023

    2023

  25. [33]

    Non- Exemplar Domain Incremental Learning via Cross-Domain Concept Integration,

    Q. Wang, Y . He, S. Dong, X. Gao, S. Wang, and Y . Gong, “Non- Exemplar Domain Incremental Learning via Cross-Domain Concept Integration,” in Europ. Conf. on Computer Vision (ECCV) , Milan, Italy, Sep. 2024, pp. 144–162

    2024

  26. [34]

    A Unified Approach to Domain Incremental Learning with Memory: Theory and Algorithm,

    H. Shi and H. Wang, “A Unified Approach to Domain Incremental Learning with Memory: Theory and Algorithm,” in Advances in Neural Information Processing Systems (NIPS) , Nov. 2023

    2023

  27. [35]

    Temporal Source Recovery for Time-Series Source-Free Unsupervised Domain Adaptation,

    Y . Wang, P. Gong, M. Wu, F. Ott, X. Li, L. Xie, and Z. Chen, “Temporal Source Recovery for Time-Series Source-Free Unsupervised Domain Adaptation,” in arXiv preprint arXiv:2409.19635 , Sep. 2024

    2024

  28. [36]

    Domain Adaptation for Time-Series Classification to Mitigate Covariate Shift,

    F. Ott, D. R ¨ugamer, L. Heublein, B. Bischl, and C. Mutschler, “Domain Adaptation for Time-Series Classification to Mitigate Covariate Shift,” in Proc. of the ACM Intl. Conf. on Multimedia (ACMMM) , Lisboa, Portugal, Oct. 2022, pp. 5934–5943

    2022

  29. [37]

    Indoor Positioning in 5G-Advanced: Challenges and Solution Toward Centimeter-Level Accuracy with Carrier Phase Enhancements,

    J. Nikonowicz, A. Mahmood, M. I. Ashraf, E. Bj ¨ornson, and M. Gidlund, “Indoor Positioning in 5G-Advanced: Challenges and Solution Toward Centimeter-Level Accuracy with Carrier Phase Enhancements,” in IEEE Wireless Communications, vol. 31(4), Aug. 2024, pp. 268–275

    2024

  30. [38]

    Improving Replay-Based Continual Semantic Segmentation with Smart Data Selection,

    T. Kalb, B. Mauthe, and J. Beyerer, “Improving Replay-Based Continual Semantic Segmentation with Smart Data Selection,” in IEEE Intl. Conf. on Intelligent Transportation Systems (ITSC) , Macau, China, Oct. 2022

    2022

  31. [39]

    Prediction Error- based Classification for Class-Incremental Learning,

    M. Zajac, T. Tuytelaars, and G. M. van de Ven, “Prediction Error- based Classification for Class-Incremental Learning,” in Intl. Conf. on Learning Representations (ICLR) , Mar. 2024

    2024

  32. [40]

    [Online]

    SciPy, “KDTree,” 2024. [Online]. Available: https://docs.scipy.org/doc/ scipy/reference/generated/scipy.spatial.KDTree.html

    2024

  33. [41]

    Indoor Positioning Technologies,

    R. Mautz, “Indoor Positioning Technologies,” in Habilitation Thesis, ETH Z ¨urich, 2012

    2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.