Pith. sign in

REVIEW 5 major objections 5 minor 42 references

Automated Manifold Learning for Reduced Order Modeling

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Classical global manifold learning methods, automatically selected on graph subsamples, recover reduced-order PDE dynamics with accuracy matching manual tuning while cutting runtime by roughly two orders of magnitude.

desk verdict A useful empirical comparison undermined by inconsistent Q-score reporting; the broad conclusion about global methods is plausible but the quantitative claims need correction. read the letter →

arxiv 2506.01741 v1 pith:AMXQJNAW submitted 2025-06-02 cs.LG

Imran Nasim , Melanie Weber This is my paper
classification cs.LG
keywords manifoldlearningreducedordermodelingspatiotemporalproximitygraphQ-scoresautomatedmodelselectionhyperparameteroptimizationdynamicalsystemsembeddings
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 proposes treating spatiotemporal observations of physical systems as nodes of a proximity graph, then applying manifold learning to embed the graph in low dimension, interpreting the embedding as a reduced-order model of the system's dynamics. It claims that with careful method selection this recovers the dynamics well, but that representation quality is highly sensitive to the algorithm's geometric assumptions and hyperparameters. To remove the human from that tuning loop, the paper introduces an automated selection pipeline that scores candidate methods and hyperparameters on small random-walk subgraphs using local and global Q-scores, then deploys the best configuration on the full graph. Across the fKdV, KS, and SG equations, synthetic static manifolds, and ERA5 climate data, the automated pipeline matches or beats manually tuned embeddings while cutting runtime by roughly two orders of magnitude.

What carries the argument

The spatial-temporal proximity graph is the central object: nodes represent time points, node attributes encode the spatial discretization, and edges connect each node to its k nearest neighbors in attribute space. Low-dimensional embeddings of this graph serve as the reduced-order dynamics. The driving objective is the Lee–Verleysen Q-score, a ranking-based measure of how well pairwise distances are preserved locally and globally by an embedding; the automation pipeline optimizes Q-scores on small random-walk subgraphs so that model selection is cheap enough to be practical.

What would settle it

Run the automated pipeline to select a method and hyperparameters on random-walk subgraphs, then compute all candidate methods' Q-scores on the full graph; if the top full-graph method is selected by the subgraph procedure only rarely, or if the subgraph-selected configuration's full-graph Q-scores fall substantially below the best full-graph configuration's, the transfer assumption fails. An even sharper test: find a spatiotemporal dataset where a local or deep method achieves the best full-graph Q-scores, which would contradict the paper's claim that global classical methods dominate across these settings.

Watch

Extended reading notes

Core claim

On the paper's own terms: encoding spatiotemporal PDE trajectory data as a k-nearest-neighbor proximity graph whose nodes are time points and whose attributes are spatial discretizations, classical global embedding methods (PCA, MDS, Isomap) recover the reduced-order dynamics better than local spectral methods (LLE, spectral embedding) and random-walk or deep graph embedding methods (DeepWalk, GCN, GraphSAGE, GAT). The authors further claim that their automated model selection, which evaluates Q-scores on random-walk subgraphs and optimizes via random search or Bayesian optimization, automatically finds these global methods with representation quality comparable to the best manually tuned configuration and computation time orders of magnitude lower.

Load-bearing premise

Q-scores computed on small random-walk subgraphs rank the candidate methods in the same order as Q-scores computed on the full graph, so that a configuration chosen on a subgraph is still the right one for the whole dataset.

Editorial extensions

If this is right

  • Practitioners can replace manual grid search in manifold-learning-based reduced-order modeling with subgraph-based automated selection, making the methods usable on large graphs.
  • The consistent advantage of global methods (PCA, MDS, Isomap) indicates that preserving global pairwise or geodesic distances is the right inductive bias for spatiotemporal PDE data.
  • The automated pipeline cuts runtime by about two orders of magnitude while retaining or improving Q-scores, enabling iterative exploration of model families and hyperparameter ranges.
  • Because the framework is unsupervised and graph-based, it transfers to any time-series data where a meaningful proximity graph can be constructed, not only PDE-generated trajectories.

Reading between the lines

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

  • The subgraph ranking assumption may fail when the full dynamics contains multiple distinct regimes, since a single random walk starting at one node may sample only one regime; a testable extension is to stratify subgraphs by dynamical regime or to anchor multiple walks.
  • The automated pipeline could be extended to output a confidence or an ensemble of selected configurations, since Q-scores on subgraphs are noisy estimators of full-graph embedding quality.
  • The observed superiority of global classical methods may be specific to the adopted graph construction (k = 20 k-NN on time-point nodes); other proximity definitions or sparser graphs could change the ranking.
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

5 major / 5 minor

Summary. The paper proposes an automated manifold learning framework for reduced order modeling of spatiotemporal dynamical systems. The authors construct spatial-temporal proximity graphs from PDE-generated and real-world data, apply a range of classical and deep-learning-based manifold learning methods, and evaluate the learned embeddings with local and global Q-scores. They report that global classical methods (PCA, MDS, Isomap) outperform local and deep methods, and that an automated pipeline based on random search or Bayesian optimization over Q-scores on random-walk subgraphs achieves comparable or better accuracy in much less time than manual model selection. The paper also includes qualitative visual comparisons against Fourier-projection ground truth.

Significance. If the quantitative results are reliable, the paper makes a useful contribution by systematically comparing manifold learning methods for reduced order modeling and by introducing an AutoML-style procedure for this setting. The qualitative visualizations and the observation that global methods capture the dynamics better than local or deep methods are valuable. However, the quantitative backbone currently contains inconsistencies (non-integer Kmax values, disagreement between Table 1 and the appendix tables, and counterexamples to the 'higher accuracy' claim) that must be resolved before the central claims can be accepted. The paper does not provide code or data for reproducibility, which amplifies these concerns.

major comments (5)
  1. [§3.2, Eq. (1), Table 1] Eq. (1) defines Kmax as an arg max over K, which must be an integer, yet Table 1 reports non-integer Kmax values such as 30.99, 156.1, and 91.9 for classical methods without stating that these are averages over seeds. This makes the reported metric ill-defined and needs clarification or correction.
  2. [Table 1 vs Appendix Tables 6-8] Table 1 and Appendix Tables 6-8 are both presented as results of the manually tuned experiments, but the Q-scores disagree substantially; for example, fKdV PCA in Table 1 has Qlocal=0.55, Qglobal=0.91, Kmax=30.99, while Table 6 reports Qlocal=0.58, Qglobal=0.93, Kmax=3.0, and ISOMAP also differs. Since the method rankings and the central claim that global classical methods perform best rest on these numbers, the discrepancy must be explained or the tables corrected.
  3. [§3.1, Model selection procedure] Section 3.1 states that the model selection procedure includes 'meta-learning of geometric priors' as a key component, but no meta-learning algorithm or experiment is described anywhere in the paper; the implemented procedures are random search and Bayesian optimization. Either implement the meta-learning component or remove the claim.
  4. [§3.1 'Subsampling procedure' and §4.3] The automation pipeline selects hyperparameters by optimizing Q-scores on random-walk subgraphs, yet reports final Q-scores on the full graph. The paper does not test whether subgraph-based rankings transfer to full-graph rankings; a single random walk may miss dynamical regimes (e.g., in chaotic KS data). Please add a validation of subgraph representativeness or explicitly state this as a limitation.
  5. [§4.3, Table 2] Table 2 contradicts the claim that automatic tuning 'generally achieves a higher accuracy': for SG, Bayesian optimization yields Qlocal=0.76, below the manual Qlocal=0.88; for Swiss roll, random search yields Qlocal=0.87, below the manual 0.99. The claim should be qualified or the table corrected.
minor comments (5)
  1. [§2.1] The cross-reference 'Apx. 3.2' for the Q-score definitions is incorrect; the definitions appear in Section 3.2.
  2. [§4.3] The text mentions 'swiss role'; this should be 'swiss roll'.
  3. [§3.2] In the co-ranking matrix definition, the symbols k and l are used both as matrix indices and as summation variables, which is confusing; please use distinct notation.
  4. [Figure 2 caption] The caption refers to the Fourier projection as 'virtual best' but the term is not defined; please clarify.
  5. [§4.2] The sentence 'This suggest that a successful approach...' contains a subject-verb agreement error; change 'suggest' to 'suggests'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found: selection and evaluation use the same Q-score metric but on disjoint subgraph and full-graph data, so the reported gains are not forced; the numerical inconsistencies and missing meta-learning component are correctness risks, not circularity.

full rationale

The paper contains no derivational chain in which an output is definitionally equal to an input. The Q-score definitions (Sec. 3.2, Eqs. 1-3) are standard external metrics cited to Lee and Verleysen (2010). The automated pipeline optimizes Q-scores on random-walk subgraphs (Sec. 3.3) and the reported accuracy in Table 2 is the Q-score of the final model run on the main graph (Appendix C), so the selection and evaluation quantities are not computed on the same data instance; a subgraph optimum is not forced to be a full-graph optimum. The same-metric reuse is a methodological weakness because subgraph-to-full-graph transfer is not validated, but it is not a definitional circularity. Self-citations (Nasim and de Sousa Almeida 2024; Nasim and Henderson 2024; Nasim and Weber 2024) appear only in contextual literature lists and are not load-bearing. No uniqueness theorem or ansatz is imported from the authors' prior work. Substantive concerns are outside circularity: Table 1 reports non-integer Kmax values for classical methods although Eq. (1) defines an argmax over integer K, and Table 1 Q-scores disagree with Appendix Tables 6-8 (e.g., fKdV PCA 0.55/0.91/30.99 vs 0.58/0.93/3.0); Section 3.1 advertises meta-learning of geometric priors but the experiments implement only random search and Bayesian optimization. These issues affect reliability of the empirical claims but do not make the reasoning circular.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim (automation improves reduced-order modeling quality and scalability) rests on the hand-chosen search spaces in Tables 4 and 5, the Q-score metric, and the assumption that random-walk subgraphs are representative. The paper contains no derivation; the 'meta-learning of geometric priors' in Section 3.1 is described but not implemented, so the framework's remaining content is a hyperparameter search. No new physical entities are introduced.

free parameters (7)
  • k (kNN graph parameter) = 20 (manual), {40,50,60,70} (search)
    Number of nearest neighbors defining the spatial-temporal proximity graph; central to the graph construction, chosen by hand in the manual baseline and as a search range in the automated framework.
  • embedding dimension ndim = 3 (manual), {3,4,5} (search)
    Dimension of the learned representation; the paper fixes 3D for visualization in manual runs and searches 3 to 5 in automation.
  • random walk length w (deep methods) = 50 (manual), {50,60,70} (search)
    Controls the random-walk similarity in DeepWalk; the paper reports that increasing walk length improved scores but increased compute.
  • subgraph walk length wsub = {80,90,100}
    Length of the random walk used to subsample the input graph for model selection; a hand-chosen range.
  • number of hidden neurons (GNNs) = 256 and 512 (manual), {128,256,512,1028} (search)
    Hidden layer widths for GCN, GraphSAGE, and GAT; chosen by hand in manual runs and searched in automation.
  • number of epochs = 1000 (manual and final full-graph runs), 100 (subgraph selection)
    Training length; 100 epochs on subgraphs for selection, 1000 for full-graph retraining.
  • learning rate and optimizer = SGD lr 1e-2 (manual); lr range [1e-4,1e-2], {Adam, SGD} (search)
    Optimizer configuration is a free choice, not dictated by the problem.
assumptions (5)
  • domain assumption Manifold hypothesis: the observed spatial-temporal data lies on or near a low-dimensional manifold.
    Stated in Section 2.1: 'we assume that data lies on or near a low-dimensional manifold'; the entire approach depends on this.
  • domain assumption kNN proximity graph with Euclidean node-attribute distances encodes the relevant similarity structure of the dynamics.
    Section 2.1: nodes are connected if k-nearest neighbors with respect to node-attribute similarity; this construction replaces true geodesic distances.
  • domain assumption Q-scores are a valid measure of representation quality for dynamics recovery.
    Section 3.2 uses Q-scores as the sole quantitative evaluation and selection criterion; no external validation of their sufficiency for reduced-order modeling quality.
  • domain assumption Fourier projection on dominant modes is a valid ground truth for the reduced-order dynamics.
    Section 4.2: ground truth is obtained via a Fourier projection on the two dominant modes; this assumes linear Fourier modes are the true latent coordinates.
  • domain assumption Random-walk subgraphs are representative of the full graph topology.
    Section 3.1: model selection is performed on subgraphs from uniform random walks; the paper does not prove or test representativeness.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Automated Manifold Learning for Reduced Order Modeling." pith.science (2026). https://pith.science/paper/AMXQJNAW

@misc{pith2026250601741,
  author       = {Pith},
  title        = {Pith review of: Automated Manifold Learning for Reduced Order Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AMXQJNAW}},
  note         = {Machine review of arXiv:2506.01741}
}
read the original abstract

The problem of identifying geometric structure in data is a cornerstone of (unsupervised) learning. As a result, Geometric Representation Learning has been widely applied across scientific and engineering domains. In this work, we investigate the use of Geometric Representation Learning for the data-driven discovery of system dynamics from spatial-temporal data. We propose to encode similarity structure in such data in a spatial-temporal proximity graph, to which we apply a range of classical and deep learning-based manifold learning approaches to learn reduced order dynamics. We observe that while manifold learning is generally capable of recovering reduced order dynamics, the quality of the learned representations varies substantially across different algorithms and hyperparameter choices. This is indicative of high sensitivity to the inherent geometric assumptions of the respective approaches and suggests a need for careful hyperparameter tuning, which can be expensive in practise. To overcome these challenges, we propose a framework for Automated Manifold Learning, which selects a manifold learning approach and corresponding hyperparameter choices based on representative subsamples of the input graph. We demonstrate that the proposed framework leads to performance gains both in scalability and in the learned representations' accuracy in capturing local and global geometric features of the underlying system dynamics.

Figures

Figures reproduced from arXiv: 2506.01741 by the authors.

Figure 1
Figure 1. Overview of our Automated Representation Learning framework. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Three-dimensional visualizations of the reduced order dynamics (top row: forced Korteweg–de Vries (fKdV), [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Left plot: Geographic distribution of 2-meter temperature at the beginning of January 2024. Right plot: Temporal [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Three-dimensional visualizations of the reduced order dynamics for the KS equation for all of the techniques [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Three-dimensional visualizations of the reduced order dynamics for the fKdV equation for all of the techniques [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Three-dimensional visualizations of the reduced order dynamics for the SG equation for all of the techniques [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 32 canonical work pages

  1. [1]

    Discovering governing equations from partial measurements with deep delay autoencoders

    Joseph Bakarji, Kathleen Champion, J Nathan Kutz, and Steven L Brunton. Discovering governing equations from partial measurements with deep delay autoencoders. Proceedings of the Royal Society A, 479 0 (2276): 0 20230422, 2023

    work page 2023

  2. [2]

    Laplacian eigenmaps for dimensionality reduction and data representation

    Mikhail Belkin and Partha Niyogi. Laplacian eigenmaps for dimensionality reduction and data representation. Neural computation, 15 0 (6): 0 1373--1396, 2003

    work page 2003

  3. [3]

    Random search for hyper-parameter optimization

    James Bergstra and Yoshua Bengio. Random search for hyper-parameter optimization. Journal of Machine Learning Research, 13 0 (10): 0 281--305, 2012

    work page 2012

  4. [4]

    Benjamin J. Binder. Steady two-dimensional free-surface flow past disturbances in an open channel: Solutions of the korteweg–de vries equation and analysis of the weakly nonlinear phase space. Fluids, 4 0 (1), 2019. ISSN 2311-5521. doi:10.3390/fluids4010024. URL https://www.mdpi.com/2311-5521/4/1/24

    work page doi:10.3390/fluids4010024 2019

  5. [5]

    Free surface flows past surfboards and sluice gates

    BJ Binder and J-M Vanden-Broeck. Free surface flows past surfboards and sluice gates. European Journal of Applied Mathematics, 16 0 (5): 0 601--619, 2005

    work page 2005

  6. [6]

    Generalized multidimensional scaling: a framework for isometry-invariant partial surface matching

    Alexander M Bronstein, Michael M Bronstein, and Ron Kimmel. Generalized multidimensional scaling: a framework for isometry-invariant partial surface matching. Proceedings of the National Academy of Sciences, 103 0 (5): 0 1168--1172, 2006

    work page 2006

  7. [7]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems

    Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the national academy of sciences, 113 0 (15): 0 3932--3937, 2016

    2016

  8. [8]

    Data-driven discovery of coordinates and governing equations

    Kathleen Champion, Bethany Lusch, J Nathan Kutz, and Steven L Brunton. Data-driven discovery of coordinates and governing equations. Proceedings of the National Academy of Sciences, 116 0 (45): 0 22445--22451, 2019

    work page 2019

Show all 42 references
  1. [9]

    An introduction to topological data analysis: Fundamental and practical aspects for data scientists

    Frédéric Chazal and Bertrand Michel. An introduction to topological data analysis: Fundamental and practical aspects for data scientists. Frontiers in Artificial Intelligence, 4, 2021

    2021

  2. [10]

    Geometric NeuralPDE ( GNPnet ) Models for Learning Dynamics

    Oluwadamilola Fasina, Smita Krishnaswamy, and Aditi Krishnapriyan. Geometric NeuralPDE ( GNPnet ) Models for Learning Dynamics . 2022

    2022

  3. [11]

    Dynamics from noisy data with extreme timing uncertainty

    Russell Fung, Ataya M Hanna, Oriol Vendrell, S Ramakrishna, Tamar Seideman, Robin Santra, and Abbas Ourmazd. Dynamics from noisy data with extreme timing uncertainty. Nature, 532 0 (7600): 0 471--475, 2016

    2016

  4. [12]

    node2vec: Scalable feature learning for networks

    Aditya Grover and Jure Leskovec. node2vec: Scalable feature learning for networks. In Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining, pages 855--864, 2016

    2016

  5. [13]

    Inductive representation learning on large graphs

    Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. Advances in neural information processing systems, 30, 2017

    2017

  6. [14]

    Exact solution of the sine-gordon equation for multiple collisions of solitons

    Ryogo Hirota. Exact solution of the sine-gordon equation for multiple collisions of solitons. Journal of the Physical Society of Japan, 33 0 (5): 0 1459--1463, 1972

    1972

  7. [15]

    Sequential model-based optimization for general algorithm configuration

    Frank Hutter, Holger H Hoos, and Kevin Leyton-Brown. Sequential model-based optimization for general algorithm configuration. In Learning and Intelligent Optimization: 5th International Conference, LION 5, Rome, Italy, January 17-21, 2011. Selected Papers 5, pages 507--523. Sp...

    2011

  8. [16]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. Semi-Supervised Classification with Graph Convolutional Networks . In Proceedings of the 5th International Conference on Learning Representations, 2017

    2017

  9. [17]

    Reconstructing phase space from pde simulations

    Michael Kirby and Dieter Armbruster. Reconstructing phase space from pde simulations. Zeitschrift f \"u r angewandte Mathematik und Physik ZAMP , 43: 0 999--1022, 1992

    1992

  10. [18]

    Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis

    Joseph B Kruskal. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika, 29 0 (1): 0 1--27, 1964

    1964

  11. [19]

    Diffusion-Induced Chaos in Reaction Systems

    Yoshiki Kuramoto. Diffusion-Induced Chaos in Reaction Systems . Progress of Theoretical Physics Supplement, 64: 0 346--367, 02 1978. ISSN 0375-9687. doi:10.1143/PTPS.64.346. URL https://doi.org/10.1143/PTPS.64.346

    1978 doi

  12. [20]

    Lee and Michel Verleysen

    John A. Lee and Michel Verleysen. Scale-independent quality criteria for dimensionality reduction. Pattern Recognition Letters, 31 0 (14): 0 2248--2257, October 2010

    2010

  13. [21]

    Identifying the latent space geometry of network models through analysis of curvature

    Shane Lubold, Arun G Chandrasekhar, and Tyler H McCormick. Identifying the latent space geometry of network models through analysis of curvature. Journal of the Royal Statistical Society Series B: Statistical Methodology, 85 0 (2): 0 240--292, 2023

    2023

  14. [22]

    Deep learning for universal linear embeddings of nonlinear dynamics

    Bethany Lusch, J Nathan Kutz, and Steven L Brunton. Deep learning for universal linear embeddings of nonlinear dynamics. Nature communications, 9 0 (1): 0 4950, 2018

    2018

  15. [23]

    Muñoz Sabater

    J. Muñoz Sabater. Era5-land hourly data from 1950 to present, 2019. URL https://doi.org/10.24381/cds.e2161bac. Accessed on 26-Jul-2024

    1950 doi

  16. [24]

    Using neural implicit flow to represent latent dynamics of canonical systems

    Imran Nasim and Joao Lucas de Sousa Almeida. Using neural implicit flow to represent latent dynamics of canonical systems. In International Conference on Scientific Computing and Machine Learning, 2024

    2024

  17. [25]

    Dynamically meaningful latent representations of dynamical systems

    Imran Nasim and Michael E Henderson. Dynamically meaningful latent representations of dynamical systems. Mathematics, 12 0 (3): 0 476, 2024

    2024

  18. [26]

    Learning reduced order dynamics via geometric representations

    Imran Nasim and Melanie Weber. Learning reduced order dynamics via geometric representations. In International Conference on Scientific Computing and Machine Learning, 2024

    2024

  19. [27]

    Otto and Clarence W

    Samuel E. Otto and Clarence W. Rowley. Linearly Recurrent Autoencoder Networks for Learning Dynamics . SIAM Journal on Applied Dynamical Systems, 18, January 2019

    2019

  20. [28]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in P ython. Journal of Machine Learnin...

    2011

  21. [29]

    Deepwalk: Online learning of social representations

    Bryan Perozzi, Rami Al-Rfou, and Steven Skiena. Deepwalk: Online learning of social representations. In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, page 701–710. Association for Computing Machinery, 2014. ISBN 9781450329569

  22. [30]

    Nonlinear dimensionality reduction by locally linear embedding

    Sam T Roweis and Lawrence K Saul. Nonlinear dimensionality reduction by locally linear embedding. science, 290 0 (5500): 0 2323--2326, 2000

    2000

  23. [31]

    Karate Club: An API Oriented Open-source Python Framework for Unsupervised Learning on Graphs

    Benedek Rozemberczki, Oliver Kiss, and Rik Sarkar. Karate Club: An API Oriented Open-source Python Framework for Unsupervised Learning on Graphs . In Proceedings of the 29th ACM International Conference on Information and Knowledge Management (CIKM '20), page 3125–3132. ACM, 2020

    2020

  24. [32]

    Normalized cuts and image segmentation

    Jianbo Shi and Jitendra Malik. Normalized cuts and image segmentation. IEEE Transactions on pattern analysis and machine intelligence, 22 0 (8): 0 888--905, 2000

    2000

  25. [33]

    Practical bayesian optimization of machine learning algorithms

    Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical bayesian optimization of machine learning algorithms. Advances in neural information processing systems, 25, 2012

    2012

  26. [34]

    A global geometric framework for nonlinear dimensionality reduction

    Joshua B Tenenbaum, Vin de Silva, and John C Langford. A global geometric framework for nonlinear dimensionality reduction. science, 290 0 (5500): 0 2319--2323, 2000

    2000

  27. [35]

    Autone: Hyperparameter optimization for massive network embedding

    Ke Tu, Peng Cui, Xiao Wang, Fei Wang, and Wenwu Zhu. Autone: Hyperparameter optimization for massive network embedding. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. ACM, 2019

    2019

  28. [36]

    Graph attention networks

    Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, Yoshua Bengio, et al. Graph attention networks. stat, 1050 0 (20): 0 10--48550, 2017

    2017

  29. [37]

    Dynamic graph cnn for learning on point clouds

    Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon. Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (tog), 38 0 (5): 0 1--12, 2019

    2019

  30. [38]

    Neighborhood growth determines geometric priors for relational representation learning

    Melanie Weber. Neighborhood growth determines geometric priors for relational representation learning. In International Conference on Artificial Intelligence and Statistics, volume 108, pages 266--276, 2020

    2020

  31. [39]

    Exponential time-differencing with embedded runge--kutta adaptive step control

    Patrick Whalen, Moysey Brio, and Jerome V Moloney. Exponential time-differencing with embedded runge--kutta adaptive step control. Journal of Computational Physics, 280: 0 579--601, 2015

    2015

  32. [40]

    Efficient automl pipeline search with matrix and tensor factorization

    Chengrun Yang, Jicong Fan, Ziyang Wu, and Madeleine Udell. Efficient automl pipeline search with matrix and tensor factorization. arXiv preprint arXiv:2006.04216, 2020

    2006 arXiv

  33. [41]

    Tabnas: Rejection sampling for neural architecture search on tabular datasets

    Chengrun Yang, Gabriel Bender, Hanxiao Liu, Pieter-Jan Kindermans, Madeleine Udell, Yifeng Lu, Quoc V Le, and Da Huang. Tabnas: Rejection sampling for neural architecture search on tabular datasets. Advances in Neural Information Processing Systems, 35: 0 11906--11917, 2022

    2022

  34. [42]

    Automated machine learning on graphs: A survey

    Ziwei Zhang, Xin Wang, and Wenwu Zhu. Automated machine learning on graphs: A survey. arXiv preprint arXiv:2103.00742, 2021

    2021 arXiv

Pith tools

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