pith. sign in

arxiv: 2604.18816 · v1 · submitted 2026-04-20 · 💻 cs.LG · cs.AI

Curvature-Aware PCA with Geodesic Tangent Space Aggregation for Semi-Supervised Learning

Alexandre L. M. Levada This is my paper

Pith reviewed 2026-05-10 05:03 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords curvature-aware PCAgeodesic tangent space aggregationsemi-supervised dimensionality reductionmanifold learninglocal covariance operatorsgraph-based alignmentspectral embeddinghigh-curvature regimes
0
0 comments X

The pith

GTSA-PCA replaces global covariance with curvature-weighted local operators on a neighbor graph and aligns the resulting tangent spaces using geodesic distances to produce geometry-aware embeddings.

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

The paper develops a geometric extension of PCA that respects the curved shape of data manifolds instead of assuming a flat linear structure. It computes local covariance matrices that are weighted according to curvature estimates drawn from a k-nearest-neighbor graph, then synchronizes these local tangent subspaces with an operator that respects shortest paths along the graph. A modest amount of label information is incorporated into the alignment step to sharpen separation between classes. This approach keeps the spectral stability of classical PCA while adapting to nonlinearity, which matters for data sets that lie on bent surfaces and are available only in small quantities. A sympathetic reader would expect the resulting embeddings to preserve both variance structure and intrinsic geometry better than either pure linear or purely nonlinear alternatives.

Core claim

GTSA-PCA replaces the single global covariance operator with curvature-weighted local covariance operators defined over a k-nearest-neighbor graph, yielding local tangent subspaces that adapt to the manifold while suppressing high-curvature distortions. A geodesic alignment operator then combines intrinsic graph distances with subspace affinities to globally synchronize these local representations. The leading components of the resulting operator define a geometry-aware embedding, and semi-supervised information is used to guide the alignment, improving discriminative structure with minimal supervision.

What carries the argument

The geodesic alignment operator that fuses intrinsic graph distances with local subspace affinities to synchronize curvature-adapted tangent spaces into a global spectral embedding.

If this is right

  • The embeddings preserve PCA's spectral stability while adapting to manifold curvature.
  • Performance gains are largest in small-sample and high-curvature regimes.
  • Minimal label information improves class separation without requiring full supervision.
  • The method supplies a unified spectral framework that links statistical variance maximization with geometric consistency.
  • It avoids some of the instability that can appear in purely nonlinear manifold techniques.

Where Pith is reading between the lines

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

  • The same local weighting and alignment idea could be applied to other linear spectral methods such as linear discriminant analysis to handle curved class boundaries.
  • Because the approach depends on a k-NN graph, performance may be sensitive to neighborhood size; multi-scale or adaptive graphs could reduce that dependence.
  • If the curvature estimates remain accurate under sampling variation, the technique could extend to time-series data whose underlying manifold changes over time.
  • Downstream tasks such as clustering or regression performed on the resulting embeddings may inherit geometric advantages without further parameter tuning.

Load-bearing premise

A k-nearest-neighbor graph plus local covariance estimates can reliably capture manifold curvature and geodesic distances without introducing uncontrolled approximation errors that distort the final embedding.

What would settle it

On a synthetic data set sampled from a known high-curvature manifold such as a Swiss roll with added noise and very few samples, the method produces embeddings whose downstream classification accuracy or reconstruction error fails to exceed that of ordinary PCA.

Figures

Figures reproduced from arXiv: 2604.18816 by Alexandre L. M. Levada.

Figure 1
Figure 1. Figure 1: Illustration of the tangent space at a point [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The shape operator measures how a normal vector changes from the tail to the tip of a tangent vector [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Scatterplots obtained after agglomerative clustering is applied to the reduced data in cardiotocography dataset. [PITH_FULL_IMAGE:figures/full_fig_p025_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Scatterplots obtained after agglomerative clustering is applied to the reduced data in the optdigits dataset. [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Scatterplots obtained after agglomerative clustering is applied to the reduced data in the Breast dataset. From [PITH_FULL_IMAGE:figures/full_fig_p026_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Scatterplots obtained after HDBSCAN clustering is applied to the reduced data in the mfeat-factors dataset. [PITH_FULL_IMAGE:figures/full_fig_p026_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Scatterplots obtained after HDBSCAN clustering is applied to the reduced data in the breast_cancer dataset. [PITH_FULL_IMAGE:figures/full_fig_p027_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Scatterplots obtained after HDBSCAN clustering is applied to the reduced data in the page-blocks dataset. [PITH_FULL_IMAGE:figures/full_fig_p027_8.png] view at source ↗
read the original abstract

Principal Component Analysis (PCA) is a fundamental tool for representation learning, but its global linear formulation fails to capture the structure of data supported on curved manifolds. In contrast, manifold learning methods model nonlinearity but often sacrifice the spectral structure and stability of PCA. We propose \emph{Geodesic Tangent Space Aggregation PCA (GTSA-PCA)}, a geometric extension of PCA that integrates curvature awareness and geodesic consistency within a unified spectral framework. Our approach replaces the global covariance operator with curvature-weighted local covariance operators defined over a $k$-nearest neighbor graph, yielding local tangent subspaces that adapt to the manifold while suppressing high-curvature distortions. We then introduce a geodesic alignment operator that combines intrinsic graph distances with subspace affinities to globally synchronize these local representations. The resulting operator admits a spectral decomposition whose leading components define a geometry-aware embedding. We further incorporate semi-supervised information to guide the alignment, improving discriminative structure with minimal supervision. Experiments on real datasets show consistent improvements over PCA, Kernel PCA, Supervised PCA and strong graph-based baselines such as UMAP, particularly in small sample size and high-curvature regimes. Our results position GTSA-PCA as a principled bridge between statistical and geometric approaches to dimensionality reduction.

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 / 1 minor

Summary. The paper proposes Geodesic Tangent Space Aggregation PCA (GTSA-PCA), a geometric extension of PCA for data on curved manifolds in a semi-supervised setting. It replaces the global covariance with curvature-weighted local covariance operators defined on a k-nearest-neighbor graph to obtain adaptive local tangent subspaces, introduces a geodesic alignment operator that combines intrinsic graph distances with subspace affinities to synchronize these local representations, and incorporates semi-supervised information to guide the alignment. The resulting spectral embedding is claimed to yield consistent empirical improvements over PCA, Kernel PCA, Supervised PCA, and graph-based baselines such as UMAP, especially in small-sample and high-curvature regimes.

Significance. If the construction can be shown to produce gains attributable to genuine manifold geometry rather than graph artifacts or the semi-supervised term, the work would offer a principled spectral bridge between linear PCA and nonlinear manifold methods, with particular value for limited-label scenarios on curved data. The emphasis on high-curvature regimes targets a recognized limitation of standard PCA.

major comments (2)
  1. [Method overview / operator construction] The curvature-weighted local covariance operators and the subsequent geodesic alignment operator are presented only at a high level with no explicit equations, derivations, consistency proofs, or approximation-error bounds. This is load-bearing for the central claim that the embedding improvements arise from curvature awareness and geodesic consistency, because (as the skeptic notes) k-NN-based local tangent estimates are sensitive to k, noise, and sampling density in high-curvature regimes; any bias propagates directly into the alignment operator and final eigenvectors.
  2. [Experiments] No ablation studies, parameter-sensitivity analysis (e.g., to k or curvature estimation), or dataset details are supplied to support the reported gains over PCA/KPCA/UMAP. Without these controls it is impossible to isolate the contribution of the proposed curvature weighting and alignment from implicit regularization or the semi-supervised term, undermining the cross-regime claim in the abstract.
minor comments (1)
  1. [Abstract] The abstract refers to 'real datasets' without naming them or providing basic statistics (dimensionality, sample size, curvature characteristics), which would aid reproducibility and assessment of the small-sample/high-curvature regime.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments, which highlight important areas for strengthening the manuscript. We address each major point below and outline planned revisions to improve clarity and rigor without altering the core claims.

read point-by-point responses

  1. Referee: [Method overview / operator construction] The curvature-weighted local covariance operators and the subsequent geodesic alignment operator are presented only at a high level with no explicit equations, derivations, consistency proofs, or approximation-error bounds. This is load-bearing for the central claim that the embedding improvements arise from curvature awareness and geodesic consistency, because k-NN-based local tangent estimates are sensitive to k, noise, and sampling density in high-curvature regimes; any bias propagates directly into the alignment operator and final eigenvectors.

    Authors: We agree that the operators are currently described at a high level in the manuscript. In the revision we will insert the explicit definitions: the curvature-weighted local covariance as a sum over k-NN neighborhoods with weights inversely proportional to local curvature estimates obtained from the graph, and the geodesic alignment operator as a product of the graph-distance kernel and the subspace-affinity matrix. A short derivation showing how the combined operator yields the final eigenvectors will be added. Full consistency proofs and approximation-error bounds are not present and would require substantial new theoretical analysis beyond the current scope; we will instead expand the discussion section with references to existing manifold-learning error analyses and additional intuition on bias mitigation. This revision directly addresses the load-bearing concern while remaining honest about theoretical limitations. revision: partial

  2. Referee: [Experiments] No ablation studies, parameter-sensitivity analysis (e.g., to k or curvature estimation), or dataset details are supplied to support the reported gains over PCA/KPCA/UMAP. Without these controls it is impossible to isolate the contribution of the proposed curvature weighting and alignment from implicit regularization or the semi-supervised term, undermining the cross-regime claim in the abstract.

    Authors: We acknowledge the absence of these controls in the submitted version. The revised manuscript will include a dedicated ablation subsection that removes the curvature-weighting term and the geodesic-alignment step independently, together with sensitivity plots varying k and the curvature-estimation hyperparameter. Expanded dataset descriptions will report sample sizes, ambient and intrinsic dimensions, and qualitative curvature indicators for each benchmark. These additions will allow readers to isolate the geometric components from the semi-supervised guidance and other regularizers, directly supporting the abstract's cross-regime claims. revision: yes

standing simulated objections not resolved
  • Rigorous consistency proofs and approximation-error bounds for the curvature-weighted operators and geodesic alignment in high-curvature regimes.

Circularity Check

0 steps flagged

No circularity: GTSA-PCA is a constructive definition validated by experiments

full rationale

The paper defines GTSA-PCA via explicit construction: curvature-weighted local covariances on a k-NN graph to obtain adaptive tangent subspaces, followed by a geodesic alignment operator combining graph distances and subspace affinities, whose spectral decomposition yields the embedding (with optional semi-supervised guidance). Claims of improvement rest on empirical results across real datasets, not on any derivation that reduces to fitted inputs, self-citations, or ansatzes. No self-definitional steps, predictions equivalent to fits, or load-bearing self-citations appear. The chain is self-contained as a proposed geometric extension of PCA.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the method implicitly relies on standard manifold assumptions and k-NN graph construction but does not enumerate them.

pith-pipeline@v0.9.0 · 5514 in / 1051 out tokens · 61816 ms · 2026-05-10T05:03:38.668795+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

166 extracted references · 166 canonical work pages

  1. [1]

    arXiv.org , year =

    Adaptive k-nearest neighbor classifier based on the local estimation of the shape operator , author =. arXiv.org , year =. doi:10.48550/arXiv.2409.05084 , url =

    work page doi:10.48550/arxiv.2409.05084

  2. [2]

    None , year =

    Adaptive k-Nearest Neighbor Classification Based on a Dynamic Number of Nearest Neighbors , author =. None , year =. doi:None , url =

    work page

  3. [3]

    None , year =

    On the Use of Modified Adaptive Nearest Neighbors for Classification , author =. None , year =. doi:10.5351/KJAS.2010.23.6.1093 , url =

    work page doi:10.5351/kjas.2010.23.6.1093 2010

  4. [4]

    ACM Symposium on Applied Computing , year =

    Adaptive kNN using expected accuracy for classification of geo-spatial data , author =. ACM Symposium on Applied Computing , year =. doi:10.1145/3167132.3167226 , url =

    work page doi:10.1145/3167132.3167226

  5. [5]

    None , year =

    An adaptive k-nearest neighbor algorithm , author =. None , year =. doi:10.1109/FSKD.2010.5569740 , url =

    work page doi:10.1109/fskd.2010.5569740 2010

  6. [6]

    None , year =

    Classification Algorithm Based on Natural Nearest Neighbor , author =. None , year =. doi:10.12733/JICS20105267 , url =

    work page doi:10.12733/jics20105267

  7. [7]

    International Journal of Intelligent Systems and Applications , year =

    An Enhanced Adaptive k-Nearest Neighbor Classifier Using Simulated Annealing , author =. International Journal of Intelligent Systems and Applications , year =. doi:10.5815/IJISA.2021.01.03 , url =

    work page doi:10.5815/ijisa.2021.01.03 2021

  8. [8]

    International Journal of Intelligent Systems , year =

    Bayes-Decisive Linear KNN with Adaptive Nearest Neighbors , author =. International Journal of Intelligent Systems , year =. doi:10.1155/2024/6664942 , url =

    work page doi:10.1155/2024/6664942 2024

  9. [9]

    ACM Cloud and Autonomic Computing Conference , year =

    Seabed sediments classification based on side-scan sonar images with adaptive feature fusion and metric learning , author =. ACM Cloud and Autonomic Computing Conference , year =. doi:10.1109/CAC57257.2022.10055663 , url =

    work page doi:10.1109/cac57257.2022.10055663 2022

  10. [10]

    International Journal of Engineering and Advanced Technology , year =

    Adaptive k-Nearest Centroid Neighbor Classifier for Detecting Drifted Twitter Spam , author =. International Journal of Engineering and Advanced Technology , year =. doi:10.35940/ijeat.e1048.0585s19 , url =

    work page doi:10.35940/ijeat.e1048.0585s19

  11. [11]

    Multidisciplinary Digital Publishing Institute , year =

    Mechanical Fault Diagnosis of High-Voltage Circuit Breakers with Dynamic Multi-Attention Graph Convolutional Networks Based on Adaptive Graph Construction , author =. Multidisciplinary Digital Publishing Institute , year =. doi:https://doi.org/10.3390/app14104036 , url =

    work page doi:10.3390/app14104036

  12. [12]

    Mathematics , year =

    A New K-Nearest Neighbors Classifier for Big Data Based on Efficient Data Pruning , author =. Mathematics , year =. doi:10.3390/math8020286 , url =

    work page doi:10.3390/math8020286

  13. [13]

    None , year =

    Partition Selection for Large-Scale Data Management Using KNN Join Processing , author =. None , year =. doi:10.1155/2020/7898230 , url =

    work page doi:10.1155/2020/7898230 2020

  14. [14]

    IEEE Transactions on Knowledge and Data Engineering , year =

    Distributed and Joint Evidential K-Nearest Neighbor Classification , author =. IEEE Transactions on Knowledge and Data Engineering , year =. doi:10.1109/TKDE.2023.3341098 , url =

    work page doi:10.1109/tkde.2023.3341098 2023

  15. [15]

    International Journal of Intelligent Decision Technologies , year =

    A Novel Nearest Neighbors Algorithm Based on Power Muirhead Mean , author =. International Journal of Intelligent Decision Technologies , year =. doi:10.48550/arXiv.2209.01514 , url =

    work page doi:10.48550/arxiv.2209.01514

  16. [16]

    IEEE Access , year =

    A Hybrid Model Based on Constraint OSELM, Adaptive Weighted SRC and KNN for Large-Scale Indoor Localization , author =. IEEE Access , year =. doi:10.1109/ACCESS.2018.2890111 , url =

    work page doi:10.1109/access.2018.2890111 2018

  17. [17]

    Institute of Electrical and Electronics Engineers , year =

    Dissimilarity Metric Learning in the Belief Function Framework , author =. Institute of Electrical and Electronics Engineers , year =. doi:https://doi.org/10.1109/tfuzz.2016.2540068 , url =

    work page doi:10.1109/tfuzz.2016.2540068 2016

  18. [18]

    Journal of Advanced Research in Applied Sciences and Engineering Technology , year =

    Recent Advancement in Prediction and Analyzation of Brain Tumour using the Artificial Intelligence Method , author =. Journal of Advanced Research in Applied Sciences and Engineering Technology , year =. doi:10.37934/araset.33.2.138150 , url =

    work page doi:10.37934/araset.33.2.138150

  19. [19]

    IEEE transactions on fuzzy systems , year =

    Active Antinoise Fuzzy Dominance Rough Feature Selection Using Adaptive K-Nearest Neighbors , author =. IEEE transactions on fuzzy systems , year =. doi:10.1109/TFUZZ.2023.3272316 , url =

    work page doi:10.1109/tfuzz.2023.3272316 2023

  20. [20]

    IEEE Access , year =

    Graph Distance and Adaptive K-Nearest Neighbors Selection-Based Density Peak Clustering , author =. IEEE Access , year =. doi:10.1109/ACCESS.2024.3403128 , url =

    work page doi:10.1109/access.2024.3403128 2024

  21. [21]

    Jurnal Gaussian , year =

    K-NEAREST NEIGHBOR DENGAN ADAPTIVE BOOSTING DAN SYNTHETIC MINORITY OVERSAMPLING TECHNIQUE UNTUK KLASIFIKASI DATA TIDAK SEIMBANG , author =. Jurnal Gaussian , year =. doi:10.14710/j.gauss.12.2.231-241 , url =

    work page doi:10.14710/j.gauss.12.2.231-241

  22. [22]

    None , year =

    Intrusion detection using k-Nearest Neighbor , author =. None , year =. doi:10.1109/ICADVC.2009.5377998 , url =

    work page doi:10.1109/icadvc.2009.5377998 2009

  23. [23]

    International Conference on Wireless Communications, Networking and Mobile Computing , year =

    Locally Adaptive Text Classification Based K-Nearest Neighbors , author =. International Conference on Wireless Communications, Networking and Mobile Computing , year =. doi:10.1109/WICOM.2007.1385 , url =

    work page doi:10.1109/wicom.2007.1385 2007

  24. [24]

    IEEE Transactions on Image Processing , year =

    An Adaptable k -Nearest Neighbors Algorithm for MMSE Image Interpolation , author =. IEEE Transactions on Image Processing , year =. doi:10.1109/TIP.2009.2023706 , url =

    work page doi:10.1109/tip.2009.2023706 2009

  25. [25]

    Open problems, questions and challenges in finite-dimensional integrable systems

    Improved Fuzzy-Optimally Weighted Nearest Neighbor Strategy to Classify Imbalanced Data , author =. Intelligent Networks and Systems Society , year =. doi:https://doi.org/10.22266/ijies2017.0430.17 , url =

    work page doi:10.22266/ijies2017.0430.17

  26. [26]

    International Journal of Advanced Computer Science and Applications , year =

    An Efficient Ensemble Algorithm for Boosting k-Nearest Neighbors Classification Performance via Feature Bagging , author =. International Journal of Advanced Computer Science and Applications , year =. doi:10.14569/ijacsa.2024.0150677 , url =

    work page doi:10.14569/ijacsa.2024.0150677 2024

  27. [27]

    None , year =

    Comparison of Naive Bayes Classifier and K-Nearest Neighbor Algorithms with Information Gain and Adaptive Boosting for Sentiment Analysis of Spotify App Reviews , author =. None , year =. doi:10.15294/rji.v2i1.68551 , url =

    work page doi:10.15294/rji.v2i1.68551

  28. [28]

    None , year =

    Adaptive k-Nearest-Neighbor Method for Process Monitoring , author =. None , year =. doi:10.1021/ACS.IECR.7B03771 , url =

    work page doi:10.1021/acs.iecr.7b03771

  29. [29]

    None , year =

    Comparative analysis between the traditional K-Nearest Neighbor and Modifications with Weight-Calculation , author =. None , year =. doi:10.1109/ISMSIT56059.2022.9932693 , url =

    work page doi:10.1109/ismsit56059.2022.9932693 2022

  30. [30]

    Neural Computation , year =

    Comparing Classification Methods for Longitudinal fMRI Studies , author =. Neural Computation , year =. doi:10.1162/NECO_a_00024 , url =

    work page doi:10.1162/neco_a_00024

  31. [31]

    None , year =

    Efficient Cluster-Based k-Nearest-Neighbor Machine Translation-Nearest-Neighbor Machine Translation , author =. None , year =. doi:https://doi.org/10.18653/v1/2022.acl-long.154 , url =

    work page doi:10.18653/v1/2022.acl-long.154 2022

  32. [32]

    None , year =

    FAST APPROXIMATE NEAREST NEIGHBORS , author =. None , year =. doi:None , url =

    work page

  33. [33]

    Science and Information Organization , year =

    An Improved K-Nearest Neighbor Algorithm for Pattern Classification , author =. Science and Information Organization , year =. doi:https://doi.org/10.14569/ijacsa.2022.0130887 , url =

    work page doi:10.14569/ijacsa.2022.0130887 2022

  34. [34]

    IEEE International Joint Conference on Neural Network , year =

    Nonnegative matrix factorization with adaptive neighbors , author =. IEEE International Joint Conference on Neural Network , year =. doi:10.1109/IJCNN.2017.7965893 , url =

    work page doi:10.1109/ijcnn.2017.7965893 2017

  35. [35]

    International Journal of Advanced Trends in Computer Science and Engineering , year =

    A Modified Adaptive Synthetic SMOTE Approach in Graduation Success Rate Classification , author =. International Journal of Advanced Trends in Computer Science and Engineering , year =. doi:10.30534/ijatcse/2019/63862019 , url =

    work page doi:10.30534/ijatcse/2019/63862019 2019

  36. [36]

    IIUM Press, International Islamic University Malaysia , year =

    KNN BASED CLASSIFICATION OF DIGITAL MODULATED SIGNALS , author =. IIUM Press, International Islamic University Malaysia , year =. doi:https://doi.org/10.31436/iiumej.v17i2.641 , url =

    work page doi:10.31436/iiumej.v17i2.641

  37. [37]

    Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics , 1248–1253doi:10.1109/SMC52423.2021.9658766

    Adaptive Global k-Nearest Neighbors for Hierarchical Classification of Data Streams* , author =. IEEE International Conference on Systems, Man and Cybernetics , year =. doi:10.1109/SMC52423.2021.9658648 , url =

    work page doi:10.1109/smc52423.2021.9658648 2021

  38. [38]

    Elsevier BV , year =

    Oversampling techniques for imbalanced data in regression , author =. Elsevier BV , year =. doi:https://doi.org/10.1016/j.eswa.2024.124118 , url =

    work page doi:10.1016/j.eswa.2024.124118 2024

  39. [39]

    Institute of Electrical and Electronics Engineers , year =

    Discriminative Adaptive Sets for Multi-Label Classification , author =. Institute of Electrical and Electronics Engineers , year =. doi:https://doi.org/10.1109/access.2020.3041763 , url =

    work page doi:10.1109/access.2020.3041763 2020

  40. [40]

    None , year =

    Improve Adaptive k-Nearest Neighbor Algorithm using Multi-threading , author =. None , year =. doi:None , url =

    work page

  41. [41]

    Emerald Publishing Limited , year =

    Activity recognition on smartphones using an AKNN based support vectors , author =. Emerald Publishing Limited , year =. doi:https://doi.org/10.1108/sr-05-2021-0157 , url =

    work page doi:10.1108/sr-05-2021-0157 2021

  42. [42]

    Improving uncertainty quantification of deep classifiers via neighborhood conformal prediction: Novel algorithm and theoretical analysis.arXiv preprint arXiv:2303.10694,

    Improving Uncertainty Quantification of Deep Classifiers via Neighborhood Conformal Prediction: Novel Algorithm and Theoretical Analysis , author =. AAAI Conference on Artificial Intelligence , year =. doi:10.48550/arXiv.2303.10694 , url =

    work page doi:10.48550/arxiv.2303.10694

  43. [43]

    Neural Information Processing Systems , year =

    An adaptive nearest neighbor rule for classification , author =. Neural Information Processing Systems , year =. doi:None , url =

    work page

  44. [44]

    IEEE Access , year =

    Unsupervised outlier detection for mixed-valued dataset based on the adaptive k-Nearest Neighbor global network , author =. IEEE Access , year =. doi:10.1109/ACCESS.2022.3161481 , url =

    work page doi:10.1109/access.2022.3161481 2022

  45. [45]

    Salmenkivi, Marko

    Urban Land Use Classification Model Fusing Multimodal Deep Features , author =. ISPRS International Journal of Geo-Information , year =. doi:10.3390/ijgi13110378 , url =

    work page doi:10.3390/ijgi13110378

  46. [46]

    None , year =

    Analysis of Public Satisfaction Levels towards Hospital Services using the K-Nearest Neighbors Method (Case Study: XYZ Regional Public Hospital) , author =. None , year =. doi:10.46729/ijstm.v4i5.1164 , url =

    work page doi:10.46729/ijstm.v4i5.1164

  47. [47]

    None , year =

    Non-Parametric Domain Adaptation for End-to-End Speech Translation , author =. None , year =. doi:https://doi.org/10.18653/v1/2022.emnlp-main.21 , url =

    work page doi:10.18653/v1/2022.emnlp-main.21 2022

  48. [48]

    None , year =

    Adaptive prototype selection algorithm for fuzzy monotonic K-nearest neighbor , author =. None , year =. doi:10.3233/jifs-236643 , url =

    work page doi:10.3233/jifs-236643

  49. [49]

    None , year =

    Selecting Adaptive Number of Nearest Neighbors in k -Nearest Neighbor Classifier Apply Diabetes Data , author =. None , year =. doi:None , url =

    work page

  50. [50]

    Measurement science and technology , year =

    An Adaptive Thresholding Approach for Open Set Fault Diagnosis , author =. Measurement science and technology , year =. doi:10.1088/1361-6501/ad9625 , url =

    work page doi:10.1088/1361-6501/ad9625

  51. [51]

    Scientific Journal of Informatics , year =

    Comparative Analysis Performance of K-Nearest Neighbor Algorithm and Adaptive Boosting on the Prediction of Non-Cash Food Aid Recipients , author =. Scientific Journal of Informatics , year =. doi:10.15294/sji.v9i2.32369 , url =

    work page doi:10.15294/sji.v9i2.32369

  52. [52]

    None , year =

    Hybrid Classification System for Uncertain Data , author =. None , year =. doi:10.1109/TSMC.2016.2622247 , url =

    work page doi:10.1109/tsmc.2016.2622247 2016

  53. [53]

    None , year =

    Operating Frequency Estimation of Slot Antenna by Using Adapted kNN Algorithm , author =. None , year =. doi:https://doi.org/10.18201/ijisae.2018637927 , url =

    work page doi:10.18201/ijisae.2018637927

  54. [54]

    IEEE India International Conference on Power Electronics , year =

    Comparative Analysis of Fault Classification in Hybrid Microgrid Using Different Machine Learning Algorithms , author =. IEEE India International Conference on Power Electronics , year =. doi:10.1109/IICPE60303.2023.10474685 , url =

    work page doi:10.1109/iicpe60303.2023.10474685 2023

  55. [55]

    Multidisciplinary Digital Publishing Institute , year =

    Mammogram Image Enhancement Techniques for Online Breast Cancer Detection and Diagnosis , author =. Multidisciplinary Digital Publishing Institute , year =. doi:https://doi.org/10.3390/s22228818 , url =

    work page doi:10.3390/s22228818

  56. [56]

    None , year =

    The Application of Adaptive Minimum Match k-Nearest Neighbors to Identify At-Risk Students in Health Professions Education , author =. None , year =. doi:10.1097/JPA.0000000000000513 , url =

    work page doi:10.1097/jpa.0000000000000513

  57. [57]

    None , year =

    Comparative Analysis Between the Traditional and Nearest Centroid Methods in the K-Nearest Neighbor Algorithm , author =. None , year =. doi:10.1109/hora55278.2022.9799808 , url =

    work page doi:10.1109/hora55278.2022.9799808 2022

  58. [58]

    Multidisciplinary Digital Publishing Institute , year =

    An RSS TransformBased WKNN for Indoor Positioning , author =. Multidisciplinary Digital Publishing Institute , year =. doi:https://doi.org/10.3390/s21175685 , url =

    work page doi:10.3390/s21175685

  59. [59]

    Institute of Electrical and Electronics Engineers , year =

    A Detection Method for Anomaly Flow in Software Defined Network , author =. Institute of Electrical and Electronics Engineers , year =. doi:https://doi.org/10.1109/access.2018.2839684 , url =

    work page doi:10.1109/access.2018.2839684 2018

  60. [60]

    None , year =

    An adaptive k-nearest neighbor graph building technique with applications to hyperspectral imagery , author =. None , year =. doi:10.1109/WNYIPW.2014.6999481 , url =

    work page doi:10.1109/wnyipw.2014.6999481 2014

  61. [61]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

    work page

  62. [62]

    and Osindero, Simon and Teh, Yee Whye , journal =

    Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

    work page

  63. [63]

    2016 , publisher=

    Deep learning , author=. 2016 , publisher=

    work page 2016

  64. [64]

    and McCallum, Andrew and Pereira, Fernando C

    Lafferty, John D. and McCallum, Andrew and Pereira, Fernando C. N. , title =. Proceedings of the Eighteenth International Conference on Machine Learning , pages =. 2001 , isbn =

    work page 2001

  65. [65]

    International Conference on Learning Representations (ICLR) , year=

    Combining Label Propagation and Simple Models Out-performs Graph Neural Networks , author=. International Conference on Learning Representations (ICLR) , year=

    work page

  66. [66]

    New Perspectives in Graph Machine Learning, NeurIPS Workshop , year=

    Posterior Label Smoothing for Node Classification , author=. New Perspectives in Graph Machine Learning, NeurIPS Workshop , year=

    work page

  67. [67]

    2020 , issue_date =

    Yu, Bengong and Fan, Zhaodi , title =. 2020 , issue_date =. doi:10.1007/s10462-019-09793-6 , journal =

    work page doi:10.1007/s10462-019-09793-6 2020

  68. [68]

    2023 , issn =

    Noise-robust graph-based semi-supervised learning with dynamic shaving label propagation , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.asoc.2023.110371 , author =

    work page doi:10.1016/j.asoc.2023.110371 2023

  69. [69]

    Graph-Based Semi-Supervised Learning: A Comprehensive Review , year=

    Song, Zixing and Yang, Xiangli and Xu, Zenglin and King, Irwin , journal=. Graph-Based Semi-Supervised Learning: A Comprehensive Review , year=

    work page

  70. [70]

    IEEE International Conference on Computer Vision (ICCV) , year=

    Conditional Random Fields as Recurrent Neural Networks , author=. IEEE International Conference on Computer Vision (ICCV) , year=

    work page

  71. [71]

    Proceedings of the Twentieth International Conference on International Conference on Machine Learning , pages =

    Zhu, Xiaojin and Ghahramani, Zoubin and Lafferty, John , title =. Proceedings of the Twentieth International Conference on International Conference on Machine Learning , pages =. 2003 , isbn =

    work page 2003

  72. [72]

    Proceedings of the International Conference on International Conference on Machine Learning , author =

    LABEL PROPAGATION WITH WEAK SUPERVISION , publisher =. Proceedings of the International Conference on International Conference on Machine Learning , author =. 2023 , location =

    work page 2023

  73. [73]

    Proceedings of the International Conference on Learning Representations , year=

    Graph Neural Networks Exponentially Lose Expressive Power for Node Classification , author=. Proceedings of the International Conference on Learning Representations , year=

    work page

  74. [74]

    Levada, A. L. M. , journal=. Minimum Information Trees for High Dimensional Data Visualization in Clustering , year=

    work page

  75. [75]

    L M , title =

    Levada, A. L M , title =. Journal of Complex Networks , volume =. 2025 , month =

    work page 2025

  76. [76]

    Reviews of Modern Physics , volume=

    The Potts model , author=. Reviews of Modern Physics , volume=. 1982 , publisher=

    work page 1982

  77. [77]

    1992 , publisher=

    Statistical Mechanics of Phase Transitions , author=. 1992 , publisher=

    work page 1992

  78. [78]

    Physical Review E , volume=

    Local resolution-limit-free Potts model for community detection , author=. Physical Review E , volume=. 2010 , publisher=

    work page 2010

  79. [79]

    Physical Review Letters , volume=

    Detecting fuzzy community structures in complex networks with a Potts model , author=. Physical Review Letters , volume=. 2004 , publisher=

    work page 2004

  80. [80]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume =

    Julian Besag , title =. Journal of the Royal Statistical Society: Series B (Methodological) , volume =. 1974 , publisher =

    work page 1974

Showing first 80 references.