Pith. sign in

REVIEW 3 major objections 7 minor 1 cited by

A User's Guide to $\texttt{KSig}$: GPU-Accelerated Computation of the Signature Kernel

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read KSig pushes signature-kernel Gram matrices past 100,000 time steps

desk verdict KSig is a genuinely useful software paper with one new algorithmic variant; the package is solid, but RFSF-TS is under-validated and the 'strong performance' claim outruns the evidence. read the letter →

arxiv 2501.07145 v2 pith:WMDBQ6H2 submitted 2025-01-13 stat.ML cs.LG

Csaba Tóth , Danilo Jr Dela Cruz , Harald Oberhauser This is my paper · ORCID
classification stat.MLcs.LG
keywords signaturekernelrandomFourierfeaturestensorsketchGPUaccelerationGrammatrixtimeseriesclassificationmethods
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 introduces KSig, a GPU-accelerated Python package for computing signature kernels on sequential data, and argues that its combination of dual (kernel-trick) and primal (random-feature) algorithms lets kernel methods scale to time series that were previously out of reach. The central new item is RFSF-TS, a tensor-sketch variant of Random Fourier Signature Features that replaces expensive tensor products with FFT-based circular convolutions of count sketches. The paper reports that this feature map matches or beats the existing tensor-random-projection variant in accuracy at lower memory cost, with time complexity $O(ML(Dd+pQ\log Q))$. If these claims hold, practitioners can compute Gram matrices for sequences of length $10^5$ or more with primal methods while exact dual methods remain usable at moderate lengths.

What carries the argument

The load-bearing object is the recursive signature feature map. The dual algorithms use the Chen-identity recursion that computes the truncated signature kernel level by level, while the primal algorithms replace the static kernel lift by random Fourier features. For RFSF-TS, the new machinery is the tensor sketch: a count-sketch-based map that sends a tensor product $x_1\otimes\cdots\otimes x_m$ to the circular convolution of independent count sketches $\mathrm{CS}_1(x_1)\star\cdots\star\mathrm{CS}_m(x_m)$, computed in $O(Q\log Q)$ time via the FFT. The paper's recursion (15) applies this sketch at every level of the signature feature recurrence, which removes the need to form or store the tensor products.

What would settle it

Measure the empirical variance of the RFSF-TS estimator against the exact truncated signature kernel for a fixed Brownian motion path while increasing truncation level $M$ and projection size $Q$; if the relative error grows much faster than $1/\sqrt{Q}$ or the memory advantage over RFSF-TRP disappears at the claimed scales, the central scalability claim is false.

Watch

Extended reading notes

Core claim

On its own terms, the paper's claim is that one unified package covers the practical range of signature-kernel computation. Exact truncated signature kernels and PDE-based untruncated kernels are implemented as dual algorithms that are quadratic in sequence length; Random Fourier Signature Features in four projection variants (DP, DP-1D, TRP, TS) form primal algorithms that are linear in sequence length. The new contribution, RFSF-TS, replaces each tensor product in the RFSF recursion with a circular convolution of independent count sketches evaluated by FFT, reducing the projection cost to $O(MLDd + MLpQ\log Q)$ instead of the $O(MLD(d+pQ))$ of tensor random projections. Empirically, RFSF-TS and RFSF-TRP give the most accurate features for a given feature-map size, RFSF-TS uses less memory than RFSF-TRP, and diagonal-projection variants achieve the lowest absolute error when memory is the binding constraint. The experiments also show that projection size $Q$ matters more than static feature size $D$ for the two tensor methods.

Load-bearing premise

The load-bearing premise is that composing independent count sketches through FFT circular convolutions inside the signature recursion preserves the unbiasedness and controlled variance of the RFSF estimator, so the accuracy seen on the benchmark carries over to real tasks.

Editorial extensions

If this is right

  • Dual algorithms (the exact signature kernel and the PDE-based kernel) can process sequence lengths around $10^3$ on a 40 GB GPU, while primal algorithms reach $L\ge 10^5$ under the same memory budget.
  • For a fixed feature-map size, RFSF-TS and RFSF-TRP give lower approximation error than the diagonal-projection variants, and RFSF-TS achieves this with a smaller memory footprint than RFSF-TRP.
  • Increasing projection size $Q$ reduces error for RFSF-TS and RFSF-TRP much more than increasing static feature size $D$, so $D$ should not be enlarged without a matching $Q$.
  • All four primal variants scale linearly in truncation level $M$, so higher-order signature information can be added without changing the asymptotic complexity.
  • The primal feature formulation lets downstream models such as linear SVM avoid the $N\times N$ Gram matrix, reducing the dependence on the number of sequences from quadratic or cubic to linear in both time and memory.

Reading between the lines

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

  • Editorial inference: the tensor-sketch identity used here is not specific to signatures, so RFSF-TS-style sketches could be dropped into other tensorized random-feature approximations, such as polynomial-kernel feature maps, wherever memory is the bottleneck.
  • Editorial inference: the accuracy benchmark compares against the truncated signature kernel with $M=5$ and $D=Q$; a natural stress test is to measure whether RFSF-TS variance stays bounded as $M$ and $Q$ grow together on longer Brownian paths, since count-sketch variance is controlled by input norms.
  • Editorial inference: if the unbiasedness of the tensor-sketch recursion extends beyond stationary RBF kernels, the approach would transfer to non-stationary static kernels and to higher embedding orders $p>1$, where the paper's complexity analysis currently focuses on $p=1$.
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 introduces KSig, a Scikit-Learn compatible Python package for GPU-accelerated computation of signature kernels on sequential data. It presents dual algorithms (the exact truncated signature kernel and the signature-PDE kernel) and primal low-rank feature algorithms (RFSF with diagonal projection, tensor random projection, and a newly proposed tensor-sketch variant, RFSF-TS). For each method the paper gives pseudocode-level recursion, complexity tables, code examples, and scalability experiments on an A100 GPU, including memory/runtime comparisons and MAPE accuracy against the exact truncated kernel. The central claims are that KSig enables Gram-matrix computation for sequences of length up to 10^5 via primal methods, that the new RFSF-TS achieves approximation quality comparable to or better than existing RFSF variants at lower memory cost, and that the package integrates cleanly with standard Scikit-Learn workflows.

Significance. If the claims hold, KSig would be a practically valuable open-source toolbox for kernel-based learning on time series, particularly because it provides GPU kernels, a dual/primal split, and a new tensor-sketch feature map that could reduce the memory footprint of random Fourier signature features. The paper's strengths are its clear code examples, plausible complexity tables, and a publicly available implementation. The exact dual algorithms are standard and well motivated. The main caveat is that the new RFSF-TS method, which is highlighted in the abstract as giving 'strong performance', is not supported by a variance analysis and is evaluated in only a single synthetic MAPE experiment without error bars or any downstream task. Thus the significance of the central new contribution is currently not fully established, although the package itself is likely to be useful to practitioners.

major comments (3)
  1. [§3.5, Eq. (15)] The paper introduces RFSF-TS as a new algorithm but does not prove or even state explicitly that the feature map in Eq. (15) provides an unbiased estimate of the truncated signature kernel. The text only says that the tensor sketch of a tensor product is again a count sketch and cites [22, Lem. 3] for the variance of a single count-sketch inner product. However, the RFSF-TS recursion aggregates O(L^m) recursively sketched tensor terms, and Section 6.3 describes an additional level-wise normalization step. The cited lemma does not cover this composed estimator. Since the abstract's 'strong performance' claim relies on the approximation quality of this new method, the paper should either provide a derivation of unbiasedness and a variance bound under the recursion, or clearly state that such a guarantee is not yet available and temper the corresponding claim.
  2. [§4.2, Figs. 3–5] The MAPE experiments constitute the only quantitative evidence for the claim that RFSF-TS 'gives strong performance compared to existing algorithms'. Fig. 3 reports a single run on a synthetic Brownian motion dataset with M=5, N=20, L=100, and no error bars, confidence intervals, or multiple random seeds. Because RFF and tensor sketches are randomized, the differences between RFSF-TS and RFSF-TRP in Fig. 3 may be within noise. The paper should either report means and standard deviations over several independent repetitions, or provide a downstream learning task where RFSF-TS is compared with RFSF-TRP. The same lack of error bars applies to Figs. 4 and 5. Without this, the assertion in the abstract is not statistically substantiated.
  3. [§3.6 and §4 (deployment experiments)] The classification examples in the paper use RFSF-TRP (PenDigits, Section 3.6) and the exact dual kernel (FingerMovements, Section 2.4), but never use the new RFSF-TS method. Consequently, the paper does not demonstrate that RFSF-TS's approximation accuracy translates to useful downstream performance. Given that the abstract highlights RFSF-TS as a central contribution, the absence of any end-to-end evaluation of this method is a significant gap. I recommend either adding a classification or regression experiment that uses RFSF-TS, or revising the abstract and conclusion to present RFSF-TS as a promising but not yet fully validated variant.
minor comments (7)
  1. [§2.1, Eq. (1)] The double-difference operator in Eq. (1) contains a typo: 'k(x_i, y_{j-1}, x_i)' should be 'k(x_i, y_{j-1})'.
  2. [§2.3] The word 'untruecated' should be 'untruncated'.
  3. [§2.4, code snippet] The variable name 'K_sig_pde' in the code differs from the earlier 'k_sig_pde' in the same snippet; please make the capitalization consistent.
  4. [§3.2] The statement 'We leave it a straightforward exercise for the reader to check that this indeed provides an unbiased approximation' is not appropriate for a central property of the proposed methods; please state the unbiasedness result explicitly or provide a short derivation.
  5. [§4.2, MAPE formula] The MAPE equation is referred to as 'eq. 4.2' but no equation number appears in the text; please number the equation or remove the reference.
  6. [§4.2, experiments] The synthetic Brownian motion experiments do not report the bandwidth of the RBF/RFF kernel or the specific random seed(s). Please include these details to improve reproducibility.
  7. [§4.2, Fig. 3] The text says 'RFSF-TS and RFSF-TRP offer lower MAPE for a given feature map size F', but at the smallest feature map sizes the RFSF-DP-1D curve appears competitive; please rephrase to avoid overstatement.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: RFSF-TS is benchmarked against the exact truncated signature kernel, an external target, and no fitted quantity is renamed as a prediction.

full rationale

The paper's central empirical claim that RFSF-TS gives strong performance is tested by MAPE against the exact truncated signature kernel kSig (Section 4.2, Fig. 3), which is defined by the dual signature-kernel recursion (2)-(3), not by the RFSF-TS estimator itself; thus the accuracy target is external to the method. The RFSF-TS feature map (Eq. 15) is a Monte Carlo construction: it applies independent count-sketch operators and RFF draws to the RFSF tensor sum (Eq. 6), and its unbiasedness follows from linearity plus the standard tensor-sketch property cited from [22] (Eqs. 13-14); no parameter is fitted to the benchmark data and then reported as a prediction. The paper does rely on the authors' prior RFSF work [35] for definitions and for the base algorithm, and cites [33] for a universality result motivating p=1, but these citations supply background methods and theorems rather than a conclusion that is equivalent to this paper's inputs; the reported scalability (Figs. 1-2, Tables 1-2) is a direct consequence of the stated complexities, not of a self-referential fit. The absence of a variance bound for RFSF-TS and the reliance on a single seedless MAPE experiment are legitimate correctness and evaluation concerns, but they are not circularity: nothing in the derivation chain defines the target quantity in terms of the estimator or renames a fitted parameter as a prediction.

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

The central claims rest on standard signature kernel theory and existing random projection and sketching results; the new tensor-sketch variant inherits these without new proofs, and the experimental conclusions depend on user-chosen hyperparameters M, p, D, and Q.

free parameters (4)
  • Truncation level M = 5 (experiments)
    Number of signature levels used in Section 4 experiments; user-specified hyperparameter.
  • Order p = 1 (experiments)
    Algebraic embedding order; experiments fix p=1 for simplicity.
  • RFF dimension D and projection size Q = D=Q=100 to 1000 in Section 4; D=200 on PenDigits
    Sizes of random feature and projection components; accuracy and runtime depend on them.
  • RBF bandwidth sigma = median heuristic per dataset
    Static kernel bandwidth set via median heuristic in deployment examples; no theory for signature kernels.
assumptions (4)
  • standard math Chen identity for signatures gives the recursion across time steps (Sec. 2.2, Eq. 3; Sec. 3.2, Eq. 7)
    Underlies all recursive signature kernel and feature computations.
  • standard math Bochner's theorem and RFF concentration guarantees for stationary kernels (Sec. 3.1)
    Justifies unbiased approximation of static kernel by random Fourier features.
  • standard math Count sketch and tensor sketch unbiasedness and variance bounds (Pham-Pagh Lemma 3, Sec. 3.5)
    Core of the new RFSF-TS algorithm; the paper cites [22] but does not prove the composed estimator.
  • domain assumption Input sequences are finite-length and the static kernel is continuous, bounded, and stationary (Sec. 3)
    Needed for RFF and signature kernel to be applicable; excludes non-stationary or unbounded kernels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A User's Guide to $\texttt{KSig}$: GPU-Accelerated Computation of the Signature Kernel." pith.science (2026). https://pith.science/paper/WMDBQ6H2

@misc{pith2026250107145,
  author       = {Pith},
  title        = {Pith review of: A User's Guide to $\textttKSig$: GPU-Accelerated Computation of the Signature Kernel},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WMDBQ6H2}},
  note         = {Machine review of arXiv:2501.07145}
}
abstract

The signature kernel is a positive definite kernel for sequential and temporal data that has become increasingly popular in machine learning applications due to powerful theoretical guarantees, strong empirical performance, and recently introduced various scalable variations. In this chapter, we give a short introduction to $\texttt{KSig}$, a $\texttt{Scikit-Learn}$ compatible Python package that implements various GPU-accelerated algorithms for computing signature kernels, and performing downstream learning tasks. We also introduce a new algorithm based on tensor sketches which gives strong performance compared to existing algorithms. The package is available at https://github.com/tgcsaba/ksig.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Branched Signature Kernel Solvers for ODEs with rough Single-Trajectory signals

    math.NA 2026-05 unverdicted novelty 7.0 of 10

    Count-sampled nested prefixes plus branched signature kernels yield a collocation ODE solver for single rough trajectories, with a universal approximation theorem and streaming updates.

Reference graph

Works this paper leans on

38 extracted references · 31 canonical work pages · cited by 1 Pith paper

  1. [1]

    The uea multivariate time series classification archive, 2018

    Anthony Bagnall, Hoang Anh Dau, Jason Lines, Michael Flynn, James Large, Aaron Bostrom, Paul Southam, and Eamonn Keogh. The uea multivariate time series classification archive, 2018. arXiv preprint arXiv:1811.00075, 2018

    arXiv 2018

  2. [2]

    Random projection in dimensionality reduction: applications to image and text data

    Ella Bingham and Heikki Mannila. Random projection in dimensionality reduction: applications to image and text data. In Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining, pages 245–250, 2001

    work page 2001

  3. [3]

    Orlicz random fourier features

    Linda Chamakh, Emmanuel Gobet, and Zolt ´an Szab ´o. Orlicz random fourier features. Journal of Machine Learning Research, 21(145):1–37, 2020

    work page 2020

  4. [4]

    Finding frequent items in data streams

    Moses Charikar, Kevin Chen, and Martin Farach-Colton. Finding frequent items in data streams. In International Colloquium on Automata, Languages, and Programming, pages 693–703. Springer, 2002

    work page 2002

  5. [5]

    Signature Moments to Characterize Laws of Stochastic Processes

    Ilya Chevyrev and Harald Oberhauser. Signature Moments to Characterize Laws of Stochastic Processes. Journal of Machine Learning Research, 23(176):1–42, 2022

    work page 2022

  6. [6]

    Experiments with Random Projection

    Sanjoy Dasgupta. Experiments with random projection. arXiv preprint arXiv:1301.3849, 2013

    work page Pith review arXiv 2013

  7. [7]

    Kernel k-means: spectral clustering and normalized cuts

    Inderjit S Dhillon, Yuqiang Guan, and Brian Kulis. Kernel k-means: spectral clustering and normalized cuts. In Proceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining , pages 551–556, 2004

    work page 2004

  8. [8]

    Time-warping invariants of multidimensional time series

    Joscha Diehl, Kurusch Ebrahimi-Fard, and Nikolas Tapia. Time-warping invariants of multidimensional time series. Acta Applicandae Mathematicae, 170(1):265–290, 2020

    work page 2020

Show all 38 references
  1. [9]

    Generalized iterated-sums signatures

    Joscha Diehl, Kurusch Ebrahimi-Fard, and Nikolas Tapia. Generalized iterated-sums signatures. Journal of Algebra, 632:801–824, 2023

    2023

  2. [10]

    On the Nystr ¨om method for approximating a gram matrix for improved kernel-based learning

    Petros Drineas, Michael W Mahoney, and Nello Cristianini. On the Nystr ¨om method for approximating a gram matrix for improved kernel-based learning. Journal of Machine Learning Research, 6(12), 2005

    2005

  3. [11]

    Improving the gaussian process sparse spectrum approximation by representing uncer- tainty in frequency inputs

    Yarin Gal and Richard Turner. Improving the gaussian process sparse spectrum approximation by representing uncer- tainty in frequency inputs. In International Conference on Machine Learning, pages 655–664, 2015

    2015

  4. [12]

    Large sample analysis of the median heuristic

    Damien Garreau, Wittawat Jitkrittum, and Motonobu Kanagawa. Large sample analysis of the median heuristic. arXiv preprint arXiv:1707.07269, 2017

    2017 arXiv

  5. [13]

    Borgwardt, Malte J

    Arthur Gretton, Karsten M. Borgwardt, Malte J. Rasch, Bernhard Sch¨olkopf, and Alexander Smola. A kernel two-sample test. J. Mach. Learn. Res., 13:723–773, March 2012

    2012

  6. [14]

    Kiraly and Harald Oberhauser

    Franz J. Kiraly and Harald Oberhauser. Kernels for sequentially ordered data. J. Mach. Learn. Res., 20(31):1–45, 2019

    2019

  7. [15]

    Numba: A llvm-based python jit compiler

    Siu Kwan Lam, Antoine Pitrou, and Stanley Seibert. Numba: A llvm-based python jit compiler. In Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC, pages 1–6, 2015

    2015

  8. [16]

    The signature kernel

    Darrick Lee and Harald Oberhauser. The signature kernel. arXiv preprint arXiv:2305.04625, 2023

    2023 arXiv

  9. [17]

    Very sparse random projections

    Ping Li, Trevor J Hastie, and Kenneth W Church. Very sparse random projections. In Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 287–296, 2006

    2006

  10. [18]

    Differential equations driven by rough paths

    Terry J Lyons, Michael Caruana, and Thierry L ´evy. Differential equations driven by rough paths. Springer, 2007

    2007

  11. [19]

    A generalised signature method for multivariate time series feature extraction

    James Morrill, Adeline Fermanian, Patrick Kidger, and Terry Lyons. A generalised signature method for multivariate time series feature extraction. arXiv preprint arXiv:2006.00873, 2020

    2006 arXiv

  12. [20]

    Cupy: A numpy-compatible library for nvidia gpu calculations

    Ryosuke Okuta, Yuya Unno, Daisuke Nishino, Shohei Hido, and Crissman Loomis. Cupy: A numpy-compatible library for nvidia gpu calculations. InProceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-first Annual Conference on Neural Information Processing...

    2017

  13. [21]

    Scikit-learn: Machine learning in python

    Fabian Pedregosa, Ga ¨el Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, et al. Scikit-learn: Machine learning in python. the Journal of machine Learning research, 12:2825–2830, 2011

    2011

  14. [22]

    Fast and scalable polynomial kernels via explicit feature maps

    Ninh Pham and Rasmus Pagh. Fast and scalable polynomial kernels via explicit feature maps. In Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 239–247, 2013

    2013

  15. [23]

    Weighted sums of random kitchen sinks: Replacing minimization with randomization in learning

    Ali Rahimi and Benjamin Recht. Weighted sums of random kitchen sinks: Replacing minimization with randomization in learning. In Advances in Neural Information Processing Systems, volume 21, 2008

    2008

  16. [24]

    Tensorized random projections

    Beheshteh Rakhshan and Guillaume Rabusseau. Tensorized random projections. In International Conference on Artificial Intelligence and Statistics, pages 3306–3316. PMLR, 2020

    2020

  17. [25]

    Machine learning in python: Main developments and technology trends in data science, machine learning, and artificial intelligence

    Sebastian Raschka, Joshua Patterson, and Corey Nolet. Machine learning in python: Main developments and technology trends in data science, machine learning, and artificial intelligence. arXiv preprint arXiv:2002.04803, 2020

    2002 arXiv

  18. [26]

    The great multivariate time series classification bake off: a review and experimental evaluation of recent algorithmic advances

    Alejandro Pasos Ruiz, Michael Flynn, James Large, Matthew Middlehurst, and Anthony Bagnall. The great multivariate time series classification bake off: a review and experimental evaluation of recent algorithmic advances. Data Mining and Knowledge Discovery, 35(2):401–449, 2021

    2021

  19. [27]

    The Signature Kernel Is the Solution of a Goursat PDE

    Cristopher Salvi, Thomas Cass, James Foster, Terry Lyons, and Weixin Yang. The Signature Kernel Is the Solution of a Goursat PDE. SIAM Journal on Mathematics of Data Science, 3(3):873–899, January 2021

    2021

  20. [28]

    Optimal rates for random fourier features

    Bharath Sriperumbudur and Zolt ´an Szab´o. Optimal rates for random fourier features. Advances in neural information processing systems, 28, 2015. A User’s Guide to KSig: GPU-Accelerated Computation of the Signature Kernel 21

    2015

  21. [29]

    Support Vector Machines

    Ingo Steinwart. Support Vector Machines. Springer, 2008

    2008

  22. [30]

    Tensor random projection for low memory dimension reduction

    Yiming Sun, Yang Guo, Joel A Tropp, and Madeleine Udell. Tensor random projection for low memory dimension reduction. arXiv preprint arXiv:2105.00105, 2021

    2021 arXiv

  23. [31]

    On kernel derivative approximation with random fourier features

    Zolt ´an Szab ´o and Bharath Sriperumbudur. On kernel derivative approximation with random fourier features. In The 22nd International Conference on Artificial Intelligence and Statistics, pages 827–836, 2019

    2019

  24. [32]

    Tslearn, a machine learning toolkit for time series data

    Romain Tavenard, Johann Faouzi, Gilles Vandewiele, Felix Divo, Guillaume Androz, Chester Holtz, Marie Payne, Roman Yurchak, Marc Rußwurm, Kushal Kolar, et al. Tslearn, a machine learning toolkit for time series data. Journal of machine learning research, 21(118):1–6, 2020

    2020

  25. [33]

    Seq2Tens: An efficient representation of sequences by low-rank tensor projections

    Csaba Toth, Patric Bonnier, and Harald Oberhauser. Seq2Tens: An efficient representation of sequences by low-rank tensor projections. In International Conference on Learning Representations, 2021

    2021

  26. [34]

    Capturing graphs with hypo-elliptic diffusions

    Csaba Toth, Darrick Lee, Celia Hacker, and Harald Oberhauser. Capturing graphs with hypo-elliptic diffusions. In Thirty-Sixth Conference on Neural Information Processing Systems, 2022

    2022

  27. [35]

    Random fourier signature features.arXiv preprint arXiv:2311.12214, 2023

    Csaba Toth, Harald Oberhauser, and Zoltan Szabo. Random fourier signature features.arXiv preprint arXiv:2311.12214, 2023

    2023 arXiv

  28. [36]

    Osborne, and Harald Oberhauser

    Csaba T ´oth, Masaki Adachi, Michael A. Osborne, and Harald Oberhauser. Learning to forget: Bayesian time series forecasting using recurrent sparse spectrum signature gaussian processes. arXiv preprint arXiv:2412.19727, 2024

    2024 arXiv

  29. [37]

    static” vector of shape𝐿𝑑 . 1 class Kernel (BaseEstimator, metaclass=ABCMeta): 2

    Christopher Williams and Matthias Seeger. Using the nystr¨om method to speed up kernel machines.Advances in neural information processing systems, 13, 2000. 6 Interface of the KSig Package The hierarchy of the KSig library is as follows. We outline the relevant files, objects,...

    2000

  30. [38]

    ""Class for computing the tensor sketch. 3 ... 4

    In particular, for a tensor x1⊗···⊗ x𝑚 ∈ (R𝑑)⊗𝑚, they implement a tensor projection operator 𝑃∈ 𝐿(( R𝑑)⊗𝑚, R𝑄), where𝑄= n components, such that the algorithm computesx1⊗···⊗ x𝑚↦→𝑃(x1⊗ ···⊗ x𝑚) without building the input tensor itself. This type also hasproject𝜓 and project out...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.