{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,1]],"date-time":"2026-07-01T08:39:39Z","timestamp":1782895179390,"version":"3.54.5"},"reference-count":63,"publisher":"Society for Industrial & Applied Mathematics (SIAM)","issue":"4","funder":[{"DOI":"10.13039\/501100000266","name":"Engineering and Physical Sciences Research Council","doi-asserted-by":"publisher","award":["EP\/S026347\/1"],"award-info":[{"award-number":["EP\/S026347\/1"]}],"id":[{"id":"10.13039\/501100000266","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100000266","name":"Engineering and Physical Sciences Research Council","doi-asserted-by":"publisher","award":["EP\/N510129\/1"],"award-info":[{"award-number":["EP\/N510129\/1"]}],"id":[{"id":"10.13039\/501100000266","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/100008885","name":"Lloyd\u2019s Register Foundation","doi-asserted-by":"publisher","award":["G0095"],"award-info":[{"award-number":["G0095"]}],"id":[{"id":"10.13039\/100008885","id-type":"DOI","asserted-by":"publisher"}]},{"name":"Office for National Statistics\/Alan Turing Institute partnership"},{"name":"Hong Kong Innovation and technology Commission"},{"DOI":"10.13039\/100013986","name":"Government of the United Kingdom","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100013986","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["SIAM J. Numer. Anal."],"published-print":{"date-parts":[[2026,8,31]]},"abstract":"<jats:p>Abstract.<\/jats:p>\n                  <jats:p>Signature kernels, inner products of path signatures, underpin several machine learning algorithms for multivariate time series analysis. For bounded variation paths, the signature kernel was recently shown to solve a Goursat PDE. However, existing PDE solvers only use increments as input data, leading to first-order approximation errors. These approaches become computationally intractable for highly oscillatory input paths, as they have to be resolved at a fine enough scale to accurately recover their signature kernel, resulting in significant time and memory complexities. In this paper, we extend the analysis to rough paths and show, leveraging the framework of smooth rough paths, that the resulting rough signature kernels can be approximated by a novel system of PDEs whose coefficients involve higher-order iterated integrals of the input rough paths. We show that this system of PDEs admits a unique solution and establish quantitative error bounds yielding a higher-order approximation to rough signature kernels.<\/jats:p>","DOI":"10.1137\/25m1764955","type":"journal-article","created":{"date-parts":[[2026,7,1]],"date-time":"2026-07-01T07:26:06Z","timestamp":1782890766000},"page":"1095-1125","source":"Crossref","is-referenced-by-count":0,"title":["Log-PDE Methods for Rough Signature Kernels"],"prefix":"10.1137","volume":"64","author":[{"ORCID":"https:\/\/orcid.org\/0009-0000-0735-829X","authenticated-orcid":true,"given":"Maud","family":"Lemercier","sequence":"first","affiliation":[{"name":"Mathematical Institute, University of Oxford, Oxford, UK."}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Terry","family":"Lyons","sequence":"additional","affiliation":[{"name":"Mathematical Institute, University of Oxford, Oxford, UK."}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Cristopher","family":"Salvi","sequence":"additional","affiliation":[{"name":"Department of Mathematics, Imperial College London, London, UK."}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"351","published-online":{"date-parts":[[2026,7,1]]},"reference":[{"key":"ref1","doi-asserted-by":"crossref","unstructured":"I. P. Arribas, C. Salvi, and L. Szpruch, Sig-SDEs model for quantitative finance, in Proceedings of the ACM International Conference on AI in Finance, 2020.","DOI":"10.1145\/3383455.3422553"},{"key":"ref2","unstructured":"B. Barancikova, Z. Huang, and C. Salvi, SigDiffusions: Score-based diffusion models for time series via log-signature embeddings, in The Thirteenth International Conference on Learning Representations."},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1016\/j.jcp.2023.112549"},{"key":"ref4","volume-title":"An Adaptive Algorithm for Rough Differential Equations","author":"Bayer C.","year":"2023"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1007\/s10013-022-00570-7"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1016\/j.aim.2020.107043"},{"key":"ref7","volume":"59","author":"Boutaib Y.","year":"2014","journal-title":"Rev. Roumaine Math. Pures Appl."},{"key":"ref8","first-page":"1","volume":"17","author":"Brouard C.","year":"2016","journal-title":"J. Mach. Learn. Res."},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1214\/23-AAP1973"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.1137\/25M1740681"},{"key":"ref11","unstructured":"T. Cass and C. Salvi, Lecture Notes on Rough Paths and Applications to Machine Learning, https:\/\/arxiv.org\/abs\/2404.06583, 2024."},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.1016\/j.jfa.2023.110261"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.2307\/1969671"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1016\/j.jcp.2021.110668"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.52202\/079017-4041"},{"key":"ref16","unstructured":"N. M. Cirone and C. Salvi, Parallelflow: Parallelizing Linear Transformers via Flow Discretization, preprint, arXiv:2504.00492, 2025."},{"key":"ref17","unstructured":"N. M. Cirone and C. Salvi, Rough Kernel Hedging, preprint, arXiv:2501.09683, 2025."},{"key":"ref18","doi-asserted-by":"crossref","unstructured":"T. Cochrane, P. Foster, V. Chhabra, M. Lemercier, T. Lyons, and C. Salvi, SK-Tree: A systematic malware detection algorithm on streaming trees via the signature kernel, in Proceedings of the International Conference on Cyber Security and Resilience (CSR), IEEE, 2021, pp. 35\u201340.","DOI":"10.1109\/CSR51186.2021.9527933"},{"key":"ref19","unstructured":"CoRoPa Computational Rough Paths, 2010, http:\/\/coropa.sourceforge.net\/."},{"key":"ref20","first-page":"156","volume":"9","author":"Drucker H.","year":"1996","journal-title":"Adv. Neural Inform. Process. Syst."},{"key":"ref21","volume-title":"Proceedings of the NeurIPS 2022 Temporal Graph Learning Workshop","author":"Dyer J.","year":"2022"},{"key":"ref22","first-page":"41","volume":"3","author":"Fermanian A.","year":"2023","journal-title":"IEEE BITS"},{"key":"ref23","first-page":"3121","volume":"34","author":"Fermanian A.","year":"2021","journal-title":"Adv. Neural Inform. Process. Syst."},{"key":"ref24","unstructured":"G. Flint and T. Lyons, Pathwise Approximation of SDEs by Coupling Piecewise Abelian Rough Paths, preprint, https:\/\/doi.org\/10.48550\/arXiv.1505.01298, 2015."},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1016\/j.jfa.2009.02.010"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.1017\/CBO9780511845079"},{"key":"ref27","first-page":"723","volume":"13","author":"Gretton A.","year":"2012","journal-title":"J. Mach. Learn. Res."},{"key":"ref28","unstructured":"M. Hoglund, E. Ferrucci, C. Hern\u00e1ndez, A. M. Gonzalez, C. Salvi, L. Sanchez-Betancourt, and Y. Zhang, A Neural RDE Approach for Continuous-Time Non-Markovian Stochastic Control Problems, ICML Workshop on New Frontiers in Learning, Control, and Dynamical Systems, 2023."},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.52202\/079017-1003"},{"key":"ref30","unstructured":"B. Horvath, M. Lemercier, C. Liu, T. Lyons, and C. Salvi, Optimal Stopping via Distribution Regression: A Higher Rank Signature Approach, preprint, arXiv:2304.01479, 2023."},{"key":"ref31","volume":"36","author":"Issa Z.","year":"2024","journal-title":"Adv. Neural Inform. Process. Syst."},{"key":"ref32","volume":"31","author":"Jacot A.","year":"2018","journal-title":"Adv. Neural Inform. Process. Syst."},{"key":"ref33","unstructured":"A. Janssen, Order Book Models, Signatures and Numerical Approximations of Rough Differential Equations, Ph.D. thesis, Oxford University, 2012."},{"key":"ref34","volume":"32","author":"Kidger P.","year":"2019","journal-title":"Adv. Neural Inform. Process. Syst."},{"key":"ref35","unstructured":"P. Kidger and T. Lyons, Signatory: Differentiable computations of the signature and logsignature transforms, on both CPU and GPU, in Proceedings of the International Conference on Learning Representations, 2020."},{"key":"ref36","first-page":"1","volume":"20","author":"Kir\u00e1ly F. J.","year":"2019","journal-title":"J. Mach. Learn. Res."},{"key":"ref37","unstructured":"J. Lee, Y. Bahri, R. Novak, S. S. Schoenholz, J. Pennington, and J. Sohl-Dickstein, Deep neural networks as Gaussian processes, in International Conference on Learning Representations, 2018."},{"key":"ref38","unstructured":"M. Lemercier, C. Salvi, T. Cass, E. V. Bonilla, T. Damoulas, and T. J. Lyons, SigGPDE: Scaling sparse Gaussian processes on sequential data, in Proceedings of the International Conference on Machine Learning, PMLR, 2021, pp. 6233\u20136242."},{"key":"ref39","unstructured":"M. Lemercier, C. Salvi, T. Damoulas, E. Bonilla, and T. Lyons, Distribution regression for sequential data, in Proceedings of the International Conference on Artificial Intelligence and Statistics, PMLR, 2021, pp. 3754\u20133762."},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-70139-4"},{"key":"ref41","unstructured":"T. Lyons, M. Caruana, and T. L\u00e9vy, Differential Equations Driven by Rough Paths, Ecole d\u2019\u00e9t\u00e9 de Probabilit\u00e9s de, Saint-Flour XXXIV, 2004, pp. 1\u201393."},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.4171\/rmi\/240"},{"key":"ref43","unstructured":"G. Manten, C. Casolo, E. Ferrucci, S. W. Mogensen, C. Salvi, and N. Kilbertus, Signature Kernel conditional independence tests in causal discovery for stochastic processes, in International Conference on Learning Representations, 2025."},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.1111\/rssb.12500"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.1093\/imanum\/20.3.423"},{"key":"ref46","unstructured":"J. Morrill, C. Salvi, P. Kidger, and J. Foster, Neural rough differential equations for long time series, in Proceedings of the International Conference on Machine Learning, PMLR, 2021, pp. 7829\u20137838."},{"key":"ref47","unstructured":"N. Muca Cirone, M. Lemercier, and C. Salvi, Neural signature Kernels as infinite-width-depth-limits of controlled ResNets, In\u00a0International Conference on Machine Learning, PMLR, 2023, pp. 25358\u201325425."},{"key":"ref48","unstructured":"L. Pacchiardi, Statistical Inference in Generative Models Using Scoring Rules, Ph.D. thesis, University of Oxford, 2022."},{"key":"ref49","unstructured":"A. Pannier and C. Salvi, A Path-Dependent PDE Solver Based on Signature Kernels, preprint, arXiv:2403.11738, 2024."},{"key":"ref50","unstructured":"J. Reizenstein and B. Graham, The iisignature Library: Efficient Calculation of Iterated-Integral Signatures and Log Signatures, preprint, arXiv:1802.08252, 2018."},{"key":"ref51","unstructured":"C. Salvi, Rough Paths, kernels, Differential Equations and an Algebra of Functions on Streams, Ph.D. thesis, University of Oxford, 2021."},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1137\/20M1366794"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.1016\/j.jalgebra.2023.07.024"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.52202\/068431-0098"},{"key":"ref55","first-page":"16635","volume":"34","author":"Salvi C.","year":"2021","journal-title":"Adv. Neural Inform. Process. Syst."},{"key":"ref56","unstructured":"P. R. Sam Morley and T. Lyons, RoughPy 0.1.0 - pypi, 2023, https:\/\/pypi.org\/project\/RoughPy\/."},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1162\/089976601750264965"},{"key":"ref58","unstructured":"D. Shmelev and C. Salvi, Sparse Signature Coefficient Recovery via Kernels, preprint, arXiv:2412.08579, 2024."},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1007\/978-0-387-77242-4"},{"key":"ref60","unstructured":"A. Tong, Signax\u00a00.2.1 - pypi, 2024, https:\/\/pypi.org\/project\/RoughPy\/."},{"key":"ref61","unstructured":"C. Toth and H. Oberhauser, Bayesian learning from sequential data using Gaussian processes with signature covariances, in Proceedings of the International Conference on Machine Learning, PMLR, 2020, pp. 9548\u20139560."},{"key":"ref62","volume-title":"Gaussian Processes for Machine Learning","volume":"2","author":"Williams C. K.","year":"2006"},{"key":"ref63","first-page":"1","volume":"23","author":"Wynne G.","year":"2022","journal-title":"J. Mach. Learn. Res."}],"container-title":["SIAM Journal on Numerical Analysis"],"original-title":[],"language":"en","deposited":{"date-parts":[[2026,7,1]],"date-time":"2026-07-01T07:27:18Z","timestamp":1782890838000},"score":1,"resource":{"primary":{"URL":"https:\/\/epubs.siam.org\/doi\/10.1137\/25M1764955"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7,1]]},"references-count":63,"journal-issue":{"issue":"4","published-print":{"date-parts":[[2026,8,31]]}},"alternative-id":["10.1137\/25M1764955"],"URL":"https:\/\/doi.org\/10.1137\/25m1764955","relation":{},"ISSN":["0036-1429","1095-7170"],"issn-type":[{"value":"0036-1429","type":"print"},{"value":"1095-7170","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,7,1]]}}}