Global and Local Topology-Aware Attention with Persistent Homology and Euler Biases for Time-Series Forecasting

Amir Saki; Usef Faghihi

arxiv: 2605.03163 · v1 · submitted 2026-05-04 · 💻 cs.LG · cs.AI

Global and Local Topology-Aware Attention with Persistent Homology and Euler Biases for Time-Series Forecasting

Usef Faghihi , Amir Saki This is my paper

Pith reviewed 2026-05-08 18:40 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords persistent homologytime-series forecastingtopology-aware attentionEuler characteristicattention mechanisminductive biasgeometric structureforecasting models

0 comments

The pith

Topology-aware attention using persistent homology and Euler biases improves time-series forecasting when geometry is predictive.

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

Scientific time series often encode predictive geometric structure such as connectivity, cycles, and nonlinear neighborhoods that standard dot-product attention does not represent. The paper adds this structure to attention logits via persistent homology features for dimensions H0-H2, anchored Euler characteristic transforms, and kernel-Hilbert channels. A validation-gated local residual applies topological corrections only when held-out validation data confirm benefit, under a strict no-leakage protocol. Experiments across lightweight attention, PatchTST, and TimeSeriesTransformer architectures on synthetic and real datasets show positive paired effects with heterogeneous magnitude, including mean relative RMSE reductions of 12.5% to 47.8% in improved units.

Core claim

The central claim is that a topology-aware attention framework augments standard attention logits with global persistent homology (H0-H2) and anchored Euler biases plus a guarded local residual for additional topological signals, yielding architecture-compatible improvements in forecasting accuracy precisely when the time series carries predictive geometric structure. This is demonstrated through matched paired comparisons under train-only calibration, validation-only selection, and test-only reporting across seven dataset units, three seeds, and three splits for 189 total paired evaluations.

What carries the argument

validation-gated residual that injects persistent homology (H0-H2) and anchored Euler characteristic transforms into attention logits

If this is right

Lightweight attention and Ridge models improve in 46 of 63 units with mean relative RMSE reduction of 12.5%.
PatchTST improves in 33 units, retains baseline in 20 units, and achieves 23.5% mean reduction.
TimeSeriesTransformer improves in 47 units with 47.8% mean reduction.
Positive paired effects appear only when geometry is predictive and vary in size across datasets and architectures.
The guarded residual ensures corrections are applied only under validation support.

Where Pith is reading between the lines

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

The heterogeneous gains across architectures suggest that topology injection may be most valuable for models whose base attention already captures some sequential order.
Similar guarded topological biases could be tested in non-forecasting attention tasks such as classification of geometric sequences.
The no-leakage protocol itself could serve as a template for evaluating other geometric or invariant-based additions in sequence models.

Load-bearing premise

Persistent homology features and Euler transforms extracted from the input time series capture genuine predictive geometric structure without introducing leakage or spurious correlations.

What would settle it

Running the identical no-leakage protocol on a new collection of time series known to contain no predictive geometric structure and observing zero or negative change in paired RMSE across the same number of units would refute the claim of positive effects when geometry is predictive.

Figures

Figures reproduced from arXiv: 2605.03163 by Amir Saki, Usef Faghihi.

**Figure 1.** Figure 1: Diagnostic summary for the S&P/FRED return-shape dataset. The current matched split view at source ↗

read the original abstract

Scientific time series often encode predictive geometric structure, including connectivity, cycles, shell-like geometry, directional changes, and nonlinear neighborhoods, that standard dot-product attention does not explicitly represent. We introduce a topology-aware attention framework that adds such structure to attention logits using persistent homology (H0-H2), anchored Euler characteristic transforms, and kernel-Hilbert channels. A validation-gated local residual captures local topological signals, including a Zeng-style local H0 component, only when held-out validation data support the correction. Exact Vietoris-Rips computations and smooth topological surrogates are evaluated under a no-leakage protocol with train-only calibration, validation-only selection, and test-only reporting. We evaluate guarded topology-aware variants across three architecture families: lightweight attention/Ridge, PatchTSTForRegression, and TimeSeriesTransformerForPrediction. Experiments include synthetic benchmarks isolating higher-order topology and real datasets covering CO2, S&P 500 return-window geometry, and NASA IMS bearing degradation. The audit uses matched paired comparisons across seven dataset units, three random seeds, and three chronological splits, giving 63 paired units per architecture and 189 paired units overall. Topology-aware models show positive paired effects when geometry is predictive, with heterogeneous magnitude across datasets and architectures. Lightweight attention/Ridge improves in 46 of 63 units, with mean relative RMSE reduction of 12.5% and paired randomization p=7.2e-4; PatchTST improves in 33 units and retains the baseline in 20 units, with 23.5% reduction and p=3.5e-5; and TimeSeriesTransformer improves in 47 units, with 47.8% reduction and p<1e-4. The results support topology as a validation-selected, architecture-compatible inductive bias.

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.

Desk Editor's Note private letter to a colleague

The paper adds persistent homology and Euler-based features to attention logits with a validation gate, reporting paired RMSE gains on some but not all forecasting setups under a no-leakage protocol.

read the letter

The central point is a topology-aware attention layer that injects H0-H2 persistent homology, anchored Euler transforms, and kernel-Hilbert channels into the logits, plus a local residual that only activates when validation data supports it. This combination is not standard in the cited forecasting literature, so the assembly itself counts as the main novelty claim. The experiments run three architectures on synthetic and real series (CO2, S&P returns, bearing data) with 63 paired units each, using train-only calibration, validation-only selection, and test-only reporting. That protocol is a clear strength; the paired randomization tests give low p-values and directional wins in most units, especially the larger reductions for the TimeSeriesTransformer variant. The work therefore supplies concrete evidence that these geometric features can help when the data actually contains the relevant structure. The gains are heterogeneous, however. PatchTST improves in only 33 of 63 units and keeps the baseline in 20, with a smaller average reduction than the other two families. Without the full methods and code it is hard to tell how much of the lift comes from the topological terms versus the extra capacity or the gating rule itself. The new constructs (anchored Euler transforms, kernel-Hilbert channels) also need precise definitions and ablation checks to show they are not just another set of hand-crafted features. Readers who already work on inductive biases for sequences or on domains where cycles and connectivity matter will find the setup useful to examine. The paper is coherent on its own terms and has enough empirical grounding to merit peer review; referees can test the leakage controls and the exact contribution of each topological component.

Referee Report

1 major / 2 minor

Summary. The paper claims that a topology-aware attention framework, incorporating persistent homology (H0-H2), anchored Euler characteristic transforms, and kernel-Hilbert channels into attention logits, combined with a validation-gated local residual, yields statistically significant RMSE reductions in time-series forecasting. This is evaluated under an explicit no-leakage protocol (train-only calibration, validation-only selection, test-only reporting) across lightweight attention/Ridge, PatchTST, and TimeSeriesTransformer architectures on synthetic benchmarks and real datasets (CO2, S&P 500, NASA IMS), with positive paired effects in the majority of 189 units and low p-values from randomization tests.

Significance. If the empirical results hold under the described protocol, the work establishes topology as a viable, architecture-compatible inductive bias for attention mechanisms in forecasting, particularly when geometric structure is predictive. The heterogeneous gains (e.g., 47.8% mean relative reduction for TimeSeriesTransformer) and use of validation gating provide a practical template for adding higher-order features without leakage, strengthening claims of robustness over standard dot-product attention.

major comments (1)

[§3] §3 (Topology-Aware Attention): The claim that anchored Euler transforms and kernel-Hilbert channels capture predictive geometry without spurious correlations relies on the validation gate; however, the paper should demonstrate that the fixed set of candidate features (H0-H2 dimensions and anchors) is chosen independently of test data, as any post-hoc expansion of this set on validation could undermine the no-leakage guarantee.

minor comments (2)

[Abstract and §5] The abstract and results section report 63 units per architecture but could explicitly tabulate the seven dataset units and three splits for reproducibility.
[§3.1] Notation for 'kernel-Hilbert channels' and 'Zeng-style local H0 component' should be defined with a brief equation or reference in the methods to aid readers unfamiliar with the specific topological constructions.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive evaluation and the constructive comment on the no-leakage protocol. We address the concern point by point below.

read point-by-point responses

Referee: [§3] §3 (Topology-Aware Attention): The claim that anchored Euler transforms and kernel-Hilbert channels capture predictive geometry without spurious correlations relies on the validation gate; however, the paper should demonstrate that the fixed set of candidate features (H0-H2 dimensions and anchors) is chosen independently of test data, as any post-hoc expansion of this set on validation could undermine the no-leakage guarantee.

Authors: We agree that explicit demonstration of independence is necessary to fully substantiate the no-leakage claim. The candidate feature set—persistent homology dimensions H0–H2 together with the anchored Euler characteristic transforms and kernel-Hilbert channels—is fixed a priori on the basis of standard topological invariants known to capture connectivity, cycles, and higher-order geometry in time series; it is not expanded, pruned, or otherwise adapted using validation data. Validation is used exclusively to gate the inclusion of the local residual correction. In the revised manuscript we will add a concise paragraph in §3 that states the candidate set is predetermined, remains constant across all splits, and is chosen independently of any empirical performance on validation or test data. This addition will make the protocol fully transparent without altering the experimental results or claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's load-bearing claims are empirical: topology-aware attention variants produce statistically significant RMSE reductions on held-out test data (63 paired units per architecture) under an explicit no-leakage protocol (train-only calibration, validation-only selection of corrections, test-only reporting). The validation-gated residual and paired randomization tests provide an external check. No self-definitional equations, fitted inputs renamed as predictions, or load-bearing self-citations appear in the derivation or results. The choice of homology dimensions is fixed in advance and gated by validation performance rather than test performance, keeping the reported gains independent of the modeling decisions.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 2 invented entities

The framework rests on the assumption that topological summaries extracted via persistent homology and Euler transforms are both computable and predictive for the target series; several weighting and gating parameters are introduced without external calibration.

free parameters (2)

topological bias weights
Scaling factors that add persistent-homology and Euler features to attention logits; these are learned or chosen during training.
validation gate threshold
Decision threshold determining when the local residual correction is applied; selected on validation data.

axioms (2)

domain assumption Persistent homology features (H0-H2) and anchored Euler transforms capture geometrically predictive structure in time series
Invoked when the authors state that scientific time series encode connectivity, cycles, and shell-like geometry that standard attention misses.
standard math The no-leakage protocol (train-only calibration, validation-only selection, test-only reporting) prevents information leakage from test data
Stated as the evaluation protocol used for all reported results.

invented entities (2)

anchored Euler characteristic transforms no independent evidence
purpose: To inject shell-like and higher-order geometric biases into attention logits
Introduced as a new component alongside persistent homology; no independent evidence outside the paper is provided.
kernel-Hilbert channels no independent evidence
purpose: To embed topological signals into the attention mechanism
New channel type proposed in the framework; independent evidence not supplied.

pith-pipeline@v0.9.0 · 5632 in / 1873 out tokens · 30273 ms · 2026-05-08T18:40:49.508374+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost (J(x)=½(x+x⁻¹)−1) washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

a topology-aware attention framework that injects such structure directly into attention logits using persistent homology (H0–H2), anchored Euler characteristic transforms, and kernel-Hilbert channels

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

28 extracted references · 3 canonical work pages

[1]

Gomez, Lukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in Neural Information Processing Systems, 2017

2017
[2]

Topological attention for time series forecasting.Advances in Neural Information Processing Systems, 34:24871–24882, 2021

Sebastian Zeng, Florian Graf, Christoph Hofer, and Roland Kwitt. Topological attention for time series forecasting.Advances in Neural Information Processing Systems, 34:24871–24882, 2021

2021
[3]

American Mathematical Society, 2010

Herbert Edelsbrunner and John Harer.Computational Topology: An Introduction. American Mathematical Society, 2010

2010
[4]

Topology and data.Bulletin of the American Mathematical Society, 46(2):255– 308, 2009

Gunnar Carlsson. Topology and data.Bulletin of the American Mathematical Society, 46(2):255– 308, 2009

2009
[5]

Statistical topological data analysis using persistence landscapes.Journal of Machine Learning Research, 16(3):77–102, 2015

Peter Bubenik. Statistical topological data analysis using persistence landscapes.Journal of Machine Learning Research, 16(3):77–102, 2015

2015
[6]

Persistence images: A stable vector representation of persistent homology.Journal of Machine Learning Research, 18(8):1–35, 2017

Henry Adams, Tegan Emerson, Michael Kirby, Rachel Neville, Chris Peterson, Patrick Shipman, Sofya Chepushtanova, Eric Hanson, Francis Motta, and Lori Ziegelmeier. Persistence images: A stable vector representation of persistent homology.Journal of Machine Learning Research, 18(8):1–35, 2017

2017
[7]

A stable multi-scale kernel for topological machine learning

Jan Reininghaus, Stefan Huber, Ulrich Bauer, and Roland Kwitt. A stable multi-scale kernel for topological machine learning. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4741–4748, 2015

2015
[8]

Katharine Turner, Sayan Mukherjee, and Doug M. Boyer. Persistent homology transform for modeling shapes and surfaces.Information and Inference, 3(4):310–344, 2014

2014
[9]

Justin Curry, Sayan Mukherjee, and Katharine Turner. How many directions determine a shape and other sufficiency results for two topological transforms.Transactions of the American Mathematical Society, Series B, 9(32):1006–1043, 2022

2022
[10]

Euler characteristic tools for topological data analysis

Olympio Hacquard and Vadim Lebovici. Euler characteristic tools for topological data analysis. Journal of Machine Learning Research, 25(240):1–39, 2024

2024
[11]

Differentiable Euler characteristic transforms for shape classifi- cation

Ernst Röell and Bastian Rieck. Differentiable Euler characteristic transforms for shape classifi- cation. InInternational Conference on Learning Representations, 2024

2024
[12]

Diss-l-ECT: Dissecting graph data with local Euler characteristic transforms

Julius von Rohrscheidt and Bastian Rieck. Diss-l-ECT: Dissecting graph data with local Euler characteristic transforms. InProceedings of the 42nd International Conference on Machine Learning, PMLR 267:61790–61809, 2025

2025
[13]

Stability and inference of the Euler characteristic transform

Lewis Marsh and David Beers. Stability and inference of the Euler characteristic transform. Discrete & Computational Geometry, 75:795–838, 2026. 19

2026
[14]

Self-attention with relative position representations

Peter Shaw, Jakob Uszkoreit, and Ashish Vaswani. Self-attention with relative position representations. InProceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2, pages 464–468, 2018

2018
[15]

Le, and Ruslan Salakhutdinov

Zihang Dai, Zhilin Yang, Yiming Yang, Jaime Carbonell, Quoc V. Le, and Ruslan Salakhutdinov. Transformer-XL: Attentive language models beyond a fixed-length context. InProceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2978–2988, 2019

2019
[16]

Smith, and Mike Lewis

Ofir Press, Noah A. Smith, and Mike Lewis. Train short, test long: Attention with linear biases enables input length extrapolation. InInternational Conference on Learning Representations, 2022

2022
[17]

Beltrami flow and neural diffusion on graphs

Yun Young Choi, Sun Woo Park, Minho Lee, and Youngho Woo. Topology-informed graph transformer.arXiv preprint arXiv:2402.02005, 2024

work page arXiv 2024
[18]

Attending to topological spaces: The cellular transformer.arXiv preprint arXiv:2405.14094, 2024

Rubén Ballester, Pablo Hernández-García, Mathilde Papillon, Claudio Battiloro, Nina Miolane, Tolga Birdal, Carles Casacuberta, Sergio Escalera, and Mustafa Hajij. Attending to topological spaces: The cellular transformer.arXiv preprint arXiv:2405.14094, 2024

work page arXiv 2024
[19]

Linear transformer topological masking with graph random features.arXiv preprint arXiv:2410.03462, 2024

Isaac Reid et al. Linear transformer topological masking with graph random features.arXiv preprint arXiv:2410.03462, 2024

work page arXiv 2024
[20]

Multiscale topology-enabled structure-to-sequence transformer for protein–ligand interaction predictions.Nature Machine Intelligence, 6:799–810, 2024

Dong Chen, Jian Liu, and Guo-Wei Wei. Multiscale topology-enabled structure-to-sequence transformer for protein–ligand interaction predictions.Nature Machine Intelligence, 6:799–810, 2024

2024
[21]

Smola.Learning with Kernels

Bernhard Schölkopf and Alexander J. Smola.Learning with Kernels. MIT Press, 2002

2002
[22]

The kernel trick for distances

Bernhard Schölkopf. The kernel trick for distances. InAdvances in Neural Information Processing Systems 13, pages 301–307, 2000

2000
[23]

Perea and John Harer

Jose A. Perea and John Harer. Sliding windows and persistence: An application of topological methods to signal analysis.Foundations of Computational Mathematics, 15(3):799–838, 2015

2015
[24]

Topological machine learning for multivariate time series.Journal of Experimental & Theoretical Artificial Intelligence, 34(2):311–326, 2022

Chengyuan Wu and Carol Anne Hargreaves. Topological machine learning for multivariate time series.Journal of Experimental & Theoretical Artificial Intelligence, 34(2):311–326, 2022

2022
[25]

Gobithaasan

Zixin Lin, Nur Fariha Syaqina Zulkepli, Mohd Shareduwan Mohd Kasihmuddin, and Rudrusamy U. Gobithaasan. CrossTopoNet: A cross-attention framework on topological latent feature space for time-series forecasting.Knowledge-Based Systems, 332:114904, 2025

2025
[26]

GUDHI user manual: Rips complex.https://gudhi.inria.fr/pytho n/latest/rips_complex_user.html

The GUDHI Project. GUDHI user manual: Rips complex.https://gudhi.inria.fr/pytho n/latest/rips_complex_user.html
[27]

Federal Reserve Bank of St. Louis. S&P 500, FRED series SP500.https://fred.stlouisfe d.org/series/SP500
[28]

IMS Bearings dataset.https://data.nasa.gov/dataset/ims-bearings

NASA Open Data. IMS Bearings dataset.https://data.nasa.gov/dataset/ims-bearings. 20

[1] [1]

Gomez, Lukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in Neural Information Processing Systems, 2017

2017

[2] [2]

Topological attention for time series forecasting.Advances in Neural Information Processing Systems, 34:24871–24882, 2021

Sebastian Zeng, Florian Graf, Christoph Hofer, and Roland Kwitt. Topological attention for time series forecasting.Advances in Neural Information Processing Systems, 34:24871–24882, 2021

2021

[3] [3]

American Mathematical Society, 2010

Herbert Edelsbrunner and John Harer.Computational Topology: An Introduction. American Mathematical Society, 2010

2010

[4] [4]

Topology and data.Bulletin of the American Mathematical Society, 46(2):255– 308, 2009

Gunnar Carlsson. Topology and data.Bulletin of the American Mathematical Society, 46(2):255– 308, 2009

2009

[5] [5]

Statistical topological data analysis using persistence landscapes.Journal of Machine Learning Research, 16(3):77–102, 2015

Peter Bubenik. Statistical topological data analysis using persistence landscapes.Journal of Machine Learning Research, 16(3):77–102, 2015

2015

[6] [6]

Persistence images: A stable vector representation of persistent homology.Journal of Machine Learning Research, 18(8):1–35, 2017

Henry Adams, Tegan Emerson, Michael Kirby, Rachel Neville, Chris Peterson, Patrick Shipman, Sofya Chepushtanova, Eric Hanson, Francis Motta, and Lori Ziegelmeier. Persistence images: A stable vector representation of persistent homology.Journal of Machine Learning Research, 18(8):1–35, 2017

2017

[7] [7]

A stable multi-scale kernel for topological machine learning

Jan Reininghaus, Stefan Huber, Ulrich Bauer, and Roland Kwitt. A stable multi-scale kernel for topological machine learning. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4741–4748, 2015

2015

[8] [8]

Katharine Turner, Sayan Mukherjee, and Doug M. Boyer. Persistent homology transform for modeling shapes and surfaces.Information and Inference, 3(4):310–344, 2014

2014

[9] [9]

Justin Curry, Sayan Mukherjee, and Katharine Turner. How many directions determine a shape and other sufficiency results for two topological transforms.Transactions of the American Mathematical Society, Series B, 9(32):1006–1043, 2022

2022

[10] [10]

Euler characteristic tools for topological data analysis

Olympio Hacquard and Vadim Lebovici. Euler characteristic tools for topological data analysis. Journal of Machine Learning Research, 25(240):1–39, 2024

2024

[11] [11]

Differentiable Euler characteristic transforms for shape classifi- cation

Ernst Röell and Bastian Rieck. Differentiable Euler characteristic transforms for shape classifi- cation. InInternational Conference on Learning Representations, 2024

2024

[12] [12]

Diss-l-ECT: Dissecting graph data with local Euler characteristic transforms

Julius von Rohrscheidt and Bastian Rieck. Diss-l-ECT: Dissecting graph data with local Euler characteristic transforms. InProceedings of the 42nd International Conference on Machine Learning, PMLR 267:61790–61809, 2025

2025

[13] [13]

Stability and inference of the Euler characteristic transform

Lewis Marsh and David Beers. Stability and inference of the Euler characteristic transform. Discrete & Computational Geometry, 75:795–838, 2026. 19

2026

[14] [14]

Self-attention with relative position representations

Peter Shaw, Jakob Uszkoreit, and Ashish Vaswani. Self-attention with relative position representations. InProceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2, pages 464–468, 2018

2018

[15] [15]

Le, and Ruslan Salakhutdinov

Zihang Dai, Zhilin Yang, Yiming Yang, Jaime Carbonell, Quoc V. Le, and Ruslan Salakhutdinov. Transformer-XL: Attentive language models beyond a fixed-length context. InProceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2978–2988, 2019

2019

[16] [16]

Smith, and Mike Lewis

Ofir Press, Noah A. Smith, and Mike Lewis. Train short, test long: Attention with linear biases enables input length extrapolation. InInternational Conference on Learning Representations, 2022

2022

[17] [17]

Beltrami flow and neural diffusion on graphs

Yun Young Choi, Sun Woo Park, Minho Lee, and Youngho Woo. Topology-informed graph transformer.arXiv preprint arXiv:2402.02005, 2024

work page arXiv 2024

[18] [18]

Attending to topological spaces: The cellular transformer.arXiv preprint arXiv:2405.14094, 2024

Rubén Ballester, Pablo Hernández-García, Mathilde Papillon, Claudio Battiloro, Nina Miolane, Tolga Birdal, Carles Casacuberta, Sergio Escalera, and Mustafa Hajij. Attending to topological spaces: The cellular transformer.arXiv preprint arXiv:2405.14094, 2024

work page arXiv 2024

[19] [19]

Linear transformer topological masking with graph random features.arXiv preprint arXiv:2410.03462, 2024

Isaac Reid et al. Linear transformer topological masking with graph random features.arXiv preprint arXiv:2410.03462, 2024

work page arXiv 2024

[20] [20]

Multiscale topology-enabled structure-to-sequence transformer for protein–ligand interaction predictions.Nature Machine Intelligence, 6:799–810, 2024

Dong Chen, Jian Liu, and Guo-Wei Wei. Multiscale topology-enabled structure-to-sequence transformer for protein–ligand interaction predictions.Nature Machine Intelligence, 6:799–810, 2024

2024

[21] [21]

Smola.Learning with Kernels

Bernhard Schölkopf and Alexander J. Smola.Learning with Kernels. MIT Press, 2002

2002

[22] [22]

The kernel trick for distances

Bernhard Schölkopf. The kernel trick for distances. InAdvances in Neural Information Processing Systems 13, pages 301–307, 2000

2000

[23] [23]

Perea and John Harer

Jose A. Perea and John Harer. Sliding windows and persistence: An application of topological methods to signal analysis.Foundations of Computational Mathematics, 15(3):799–838, 2015

2015

[24] [24]

Topological machine learning for multivariate time series.Journal of Experimental & Theoretical Artificial Intelligence, 34(2):311–326, 2022

Chengyuan Wu and Carol Anne Hargreaves. Topological machine learning for multivariate time series.Journal of Experimental & Theoretical Artificial Intelligence, 34(2):311–326, 2022

2022

[25] [25]

Gobithaasan

Zixin Lin, Nur Fariha Syaqina Zulkepli, Mohd Shareduwan Mohd Kasihmuddin, and Rudrusamy U. Gobithaasan. CrossTopoNet: A cross-attention framework on topological latent feature space for time-series forecasting.Knowledge-Based Systems, 332:114904, 2025

2025

[26] [26]

GUDHI user manual: Rips complex.https://gudhi.inria.fr/pytho n/latest/rips_complex_user.html

The GUDHI Project. GUDHI user manual: Rips complex.https://gudhi.inria.fr/pytho n/latest/rips_complex_user.html

[27] [27]

Federal Reserve Bank of St. Louis. S&P 500, FRED series SP500.https://fred.stlouisfe d.org/series/SP500

[28] [28]

IMS Bearings dataset.https://data.nasa.gov/dataset/ims-bearings

NASA Open Data. IMS Bearings dataset.https://data.nasa.gov/dataset/ims-bearings. 20