Geodesics of Dynamic Graphs for Regime Change Detection

\'Etienne Voutaz; Pascal Frossard; William Cappelletti

arxiv: 2606.07151 · v1 · pith:XRZYXMQJnew · submitted 2026-06-05 · 💻 cs.LG

Geodesics of Dynamic Graphs for Regime Change Detection

William Cappelletti , \'Etienne Voutaz , Pascal Frossard This is my paper

Pith reviewed 2026-06-27 22:12 UTC · model grok-4.3

classification 💻 cs.LG

keywords dynamic graphsregime change detectiongeodesicstemporal networkschange point detectiongraph regressionnetwork evolutioncontinuous dynamics

0 comments

The pith

Regimes in dynamic graphs are coherent trajectories along geodesics in graph space, with changes detected as cumulative drifts from those paths.

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

The paper claims that standard change point detection fails for most real networks because it assumes sudden jumps between fixed states, whereas actual regimes are stretches of steady evolution that can be modeled as geodesic paths connecting observed graphs. By estimating those geodesics and tracking how far a sequence of graphs wanders from the path using regression, the approach flags either a switch to a new trajectory or a change in how fast the system is moving along it. Experiments on synthetic dynamic networks with controlled trajectory shifts and speed variations show the method locates these transitions more accurately than prior detectors, and the same procedure applied to mobility graphs during the pandemic produces change points that line up more closely with documented external events.

Core claim

Regimes are defined as periods of coherent dynamics in temporal graphs and characterized as trajectories along geodesics in a suitably defined graph space; regime changes are then significant drifts in dynamics, either toward new trajectories or with pace changes, measured as the cumulative distance of observed graph sequences from the estimated geodesics between their endpoints via graph regression, which is then combined with change point detection algorithms.

What carries the argument

Geodesics in graph space, representing the shortest coherent evolution paths between endpoint graphs, with cumulative deviation from these paths quantified through graph regression to identify regime shifts.

If this is right

On synthetic dynamic networks that include both trajectory changes and speed variations, the geodesic-distance approach locates transitions more accurately than existing change point detection models.
When applied to mobility networks recorded during the Covid-19 pandemic, the detected change points align more closely with documented external events than those returned by baseline methods.
The framework supplies the first explicit model for detecting transitions between continuously evolving regimes rather than between stationary states.

Where Pith is reading between the lines

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

The same distance-to-geodesic signal could be used to forecast the most likely continuation of an ongoing regime before a change is declared.
If graph regression reliably recovers the underlying geodesic, the method might extend to ranking candidate future graphs by how far they would lie from the current trajectory.
The approach could be tested on other continuously evolving systems such as citation networks or protein-interaction graphs to check whether the geodesic assumption holds outside mobility data.

Load-bearing premise

Coherent dynamics in temporal graphs correspond to trajectories along geodesics in a suitably defined graph space.

What would settle it

A simulation of temporal graphs whose evolution follows continuous but non-geodesic paths, where the method fails to recover the known change locations better than baseline detectors.

Figures

Figures reproduced from arXiv: 2606.07151 by \'Etienne Voutaz, Pascal Frossard, William Cappelletti.

**Figure 1.** Figure 1: Illustration of the parallelism between coherent evolution in Euclidean and graph spaces, where the underlying regimes correspond respectively to straight lines and geodesic curves. Blue arrows are aligned, and they correspond to a same regime, while the red arrow shows an alternative regime that arises at change points t = 4 and t = 3 respectively. Colored points show perturbed observation, in the relati… view at source ↗

**Figure 2.** Figure 2: Example of graphs on the Linear (L2,2) and Bures-Wasserstein geodesics between two graphs G0, G1, respectively sampled from a Barabasi-Albert model, and from a SBM with two blocks. Figure 2a shows the adjacency matrices for τ = 0, 0.25, 0.5, 0.75, 1, while Figure 2b shows the evolution of each edge weight along the geodesics. With a linear geodesic all edges change independently and at the same speed (oran… view at source ↗

**Figure 3.** Figure 3: Mean and std of distances of discrete samples from the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

read the original abstract

Traditional change point detection in dynamic networks assumes abrupt transitions between stationary states, overlooking scenarios of continuous evolution which arise in most real-world applications, such as social networks or physical systems. We address this gap by formally defining regimes as periods of coherent dynamics in temporal graphs, which we characterize as trajectories along geodesics in a suitably defined graph space. This original perspective allows us to define regime changes as significant drifts in dynamics, either toward new trajectories or with pace changes. We leverage graph regression methods to measure the cumulative distance of sequences of observed graphs from the estimated geodesics between their endpoints, in the relevant graph space, which we can combine with change point detection algorithms. We present experiments on dynamic networks, with changing trajectories and varying speeds, in which we outperform state of the art change point detection models. Then, we analyse mobility data during the Covid-19 pandemic, and show that our assumptions on regular network evolution lead to change points that are more aligned to external events compared to the outcomes of baseline methods. Our work is the first to model and detect changes between evolving regimes in graph space, providing a realistic and powerful tool for analyzing complex temporal graph data.

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 reframes regime changes in dynamic graphs as drifts from geodesic trajectories to handle continuous evolution, with experiments showing gains over abrupt-transition baselines but resting on a key modeling choice.

read the letter

The core idea here is treating regimes in temporal graphs as stretches of coherent dynamics that follow geodesics in a chosen graph space, so changes show up as cumulative distance from the estimated geodesic between sequence endpoints. They estimate that geodesic, run graph regression to get a drift signal, and plug it into existing change-point detectors. This is positioned as the first explicit treatment of evolving regimes rather than just jumps between stationary states.

What stands out is the synthetic tests with varying trajectories and speeds, where the method beats standard change-point approaches, plus the Covid mobility example where detected points line up more closely with known external events. That gives some concrete evidence the framing can be useful on real gradual-shift data.

The soft spot is the central modeling assumption that coherent dynamics map to geodesic paths in the right space; this is asserted rather than derived from first principles, and the abstract gives little on how the space or metric is picked or validated. Without the full methods it is also unclear how much the regression distance depends on the same sequences used to fit the geodesics, though they combine it with off-the-shelf detectors. A fuller literature comparison would also help nail down the "first" claim.

This is aimed at researchers working on dynamic networks and change detection who already handle graph sequences and want something beyond abrupt-switch models. The empirical results are there to evaluate, so it deserves a serious referee even if the methods need tightening for reproducibility.

Referee Report

3 major / 2 minor

Summary. The paper claims that traditional change point detection overlooks continuous evolution in dynamic networks. It defines regimes as periods of coherent dynamics characterized as trajectories along geodesics in a suitably metrized graph space. Regime changes are then defined as significant drifts toward new trajectories or pace changes, measured via cumulative distance from estimated geodesics using graph regression methods applied to sequences of observed graphs. This signal is combined with standard change point detection algorithms. Experiments on synthetic dynamic networks with varying trajectories and speeds show outperformance over state-of-the-art baselines, while analysis of Covid-19 mobility data yields change points more aligned with external events.

Significance. If the central modeling perspective and empirical results hold, the work offers a geometrically grounded approach to detecting changes between evolving regimes rather than assuming abrupt stationary transitions, which aligns better with many real-world temporal graph applications in social and physical systems. The explicit use of geodesics and regression-based deviation provides a falsifiable, extensible framework whose value is demonstrated through both controlled synthetic cases and real-data alignment with known events.

major comments (3)

[Methods section (graph space and geodesic estimation)] The core claim that coherent dynamics correspond to geodesic trajectories (and that deviations from them detect regime changes) rests on the construction of the graph space and the geodesic estimation procedure. The manuscript provides only high-level definitions without explicit details on the metric, embedding, or optimization used to compute geodesics between endpoints, which is load-bearing for all subsequent claims about distance signals and change detection.
[Methods section (graph regression and distance computation)] Graph regression is used both to estimate the geodesics between sequence endpoints and to compute the cumulative distance of the same observed sequence from those geodesics. This creates a potential dependence between the fitted quantities and the deviation signal fed to the change point algorithm; the paper must demonstrate (e.g., via controlled ablation or synthetic counter-examples) that the procedure does not introduce spurious detections or circular bias.
[Experiments section] The experimental claims of outperformance and better alignment with external events rely on specific choices of change point algorithms, parameter settings, and evaluation metrics, yet the manuscript supplies insufficient detail on statistical controls, multiple-run variability, or error analysis to assess whether the reported gains are robust.

minor comments (2)

Notation for the graph space, geodesic, and regression objective should be introduced with explicit symbols and referenced consistently in the text and figures.
Figure captions and table legends should explicitly state the graph space metric and regression method used so that results can be interpreted without returning to the main text.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our geometric approach to regime change detection in dynamic graphs. We address each major comment below and will revise the manuscript accordingly to improve clarity and rigor.

read point-by-point responses

Referee: The core claim that coherent dynamics correspond to geodesic trajectories rests on the construction of the graph space and the geodesic estimation procedure. The manuscript provides only high-level definitions without explicit details on the metric, embedding, or optimization used to compute geodesics between endpoints.

Authors: We agree that the current version presents the graph space construction at a high level. In the revised manuscript we will add explicit definitions of the metric (including its properties), the embedding into the graph space, and the optimization procedure for geodesic computation between sequence endpoints. This will ensure the distance signal derivation is fully specified and reproducible. revision: yes
Referee: Graph regression is used both to estimate the geodesics between sequence endpoints and to compute the cumulative distance of the same observed sequence from those geodesics. This creates a potential dependence between the fitted quantities and the deviation signal; the paper must demonstrate that the procedure does not introduce spurious detections or circular bias.

Authors: The concern is valid. The geodesic is fitted solely between the sequence endpoints while the cumulative distance quantifies deviations of intermediate observations; this structure is not inherently circular. Nevertheless, to demonstrate absence of spurious bias we will add, in revision, a controlled ablation on synthetic trajectories that isolates the regression step and reports change-point detection performance with and without it, together with counter-examples where the signal correctly remains flat under stationary dynamics. revision: partial
Referee: The experimental claims rely on specific choices of change point algorithms, parameter settings, and evaluation metrics, yet the manuscript supplies insufficient detail on statistical controls, multiple-run variability, or error analysis to assess whether the reported gains are robust.

Authors: We concur that robustness details are currently insufficient. The revised experiments section will include: explicit parameter-selection protocol and sensitivity analysis, aggregated results over multiple independent runs with standard deviations and confidence intervals, and statistical significance tests comparing our method against baselines on both synthetic and COVID mobility data. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper introduces a modeling perspective that defines regimes as geodesic trajectories in a metrized graph space and then applies graph regression to compute cumulative distances for change-point detection. No equations, self-citations, or fitted-parameter renamings appear in the provided abstract or context that would reduce any claimed prediction or first-principles result to its own inputs by construction. The central contribution is presented as an empirical modeling choice validated on synthetic and real data rather than a theorem whose validity depends on internal self-reference.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests primarily on a domain assumption about the nature of coherent dynamics; no free parameters or invented entities are described in the abstract.

axioms (1)

domain assumption Regimes are periods of coherent dynamics in temporal graphs that can be characterized as trajectories along geodesics in a suitably defined graph space.
This definition is introduced to address the gap in handling continuous evolution and underpins the regime change detection method.

pith-pipeline@v0.9.1-grok · 5735 in / 1270 out tokens · 27914 ms · 2026-06-27T22:12:21.215852+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 13 canonical work pages · 1 internal anchor

[1]

Anomaly detection in dynamic networks: A survey,

S. Ranshous, S. Shen, D. Koutra, S. Harenberg, C. Faloutsos, and N. F. Samatova, “Anomaly detection in dynamic networks: A survey,”WIREs Computational Statistics, vol. 7, no. 3, 2015.DOI: 10.1002/wics.1347

work page doi:10.1002/wics.1347 2015
[2]

A Survey of Change Point Detection in Dynamic Graphs,

Y . Zhou, S. Gao, D. Guo, X. Wei, J. Rokne, and H. Wang, “A Survey of Change Point Detection in Dynamic Graphs,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 3, Mar. 2025.DOI: 10.1109/TKDE.2024.3523857

work page doi:10.1109/tkde.2024.3523857 2025
[3]

Selective review of offline change point detection methods,

C. Truong, L. Oudre, and N. Vayatis, “Selective review of offline change point detection methods,”Signal Pro- cessing, vol. 167, Feb. 2020.DOI: 10.1016/j.sigpro. 2019.107299

work page doi:10.1016/j.sigpro 2020
[4]

Transfer Graph Neural Networks for Pandemic Forecasting,

G. Panagopoulos, G. Nikolentzos, and M. Vazirgian- nis, “Transfer Graph Neural Networks for Pandemic Forecasting,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 6, May 2021.DOI: 10.1609/aaai.v35i6.16616

work page doi:10.1609/aaai.v35i6.16616 2021
[5]

A pruned dynamic programming algorithm to recover the best segmentations with 1 to K max change-points.,

G. Rigaill, “A pruned dynamic programming algorithm to recover the best segmentations with 1 to K max change-points.,”Journal de la Soci ´et´e Franc ¸aise de Statistique, vol. 156, no. 4, Dec. 2015

2015
[6]

A Cluster Analysis Method for Grouping Means in the Analysis of Variance,

A. J. Scott and M. Knott, “A Cluster Analysis Method for Grouping Means in the Analysis of Variance,” Biometrics, vol. 30, no. 3, 1974.DOI: 10.2307/2529204

work page doi:10.2307/2529204 1974
[7]

Optimal De- tection of Changepoints With a Linear Computational Cost,

R. Killick, P. Fearnhead, and I. A. Eckley, “Optimal De- tection of Changepoints With a Linear Computational Cost,”Journal of the American Statistical Association, vol. 107, no. 500, Dec. 2012.DOI: 10.1080/01621459. 2012.737745 10

work page doi:10.1080/01621459 2012
[8]

Concept Drift and Anomaly Detection in Graph Streams,

D. Zambon, C. Alippi, and L. Livi, “Concept Drift and Anomaly Detection in Graph Streams,”IEEE Trans- actions on Neural Networks and Learning Systems, vol. 29, no. 11, Nov. 2018.DOI: 10 . 1109 / TNNLS . 2018.2804443

arXiv 2018
[9]

R. P. Duin et al.,The Dissimilarity Representation for Pattern Recognition: Foundations and Applications. World scientific, 2005, vol. 64

2005
[10]

T. Chen, Z. Lubberts, A. Athreya, Y . Park, and C. E. Priebe,Euclidean mirrors and first-order changepoints in network time series, Oct. 2024.DOI: 10.48550/arXiv. 2405.11111

work page internal anchor Pith review doi:10.48550/arxiv 2024
[11]

A Comprehensive Survey on Graph Anomaly Detection With Deep Learning,

X. Ma et al., “A Comprehensive Survey on Graph Anomaly Detection With Deep Learning,”IEEE Trans- actions on Knowledge and Data Engineering, vol. 35, no. 12, Dec. 2023.DOI: 10.1109/TKDE.2021.3118815

work page doi:10.1109/tkde.2021.3118815 2023
[12]

Statis- tical inference for Bures–Wasserstein barycenters,

A. Kroshnin, V . Spokoiny, and A. Suvorikova, “Statis- tical inference for Bures–Wasserstein barycenters,”The Annals of Applied Probability, vol. 31, no. 3, Jun. 2021. DOI: 10.1214/20-AAP1618

work page doi:10.1214/20-aap1618 2021
[13]

On the Bures– Wasserstein distance between positive definite matri- ces,

R. Bhatia, T. Jain, and Y . Lim, “On the Bures– Wasserstein distance between positive definite matri- ces,”Expositiones Mathematicae, vol. 37, no. 2, Jun. 2019.DOI: 10.1016/j.exmath.2018.01.002

work page doi:10.1016/j.exmath.2018.01.002 2019
[14]

GOT: An Optimal Transport framework for Graph comparison,

H. Petric Maretic, M. El Gheche, G. Chierchia, and P. Frossard, “GOT: An Optimal Transport framework for Graph comparison,” inAdvances in Neural Information Processing Systems, vol. 32, Curran Associates, Inc., 2019

2019
[15]

Bures-Wasserstein Means of Graphs,

I. Haasler and P. Frossard, “Bures-Wasserstein Means of Graphs,” inProceedings of The 27th International Con- ference on Artificial Intelligence and Statistics, PMLR, Apr. 2024

2024
[16]

At- tribute Fusion in a Latent Process Model for Time Series of Graphs,

M. Tang, Y . Park, N. H. Lee, and C. E. Priebe, “At- tribute Fusion in a Latent Process Model for Time Series of Graphs,”IEEE Transactions on Signal Processing, vol. 61, no. 7, Apr. 2013.DOI: 10 . 1109 / TSP. 2013 . 2243445

2013
[17]

Laplacian Change Point Detection for Dynamic Graphs,

S. Huang, Y . Hitti, G. Rabusseau, and R. Rab- bany, “Laplacian Change Point Detection for Dynamic Graphs,” inProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Virtual Event CA USA: ACM, Aug. 2020. DOI: 10.1145/3394486.3403077

work page doi:10.1145/3394486.3403077 2020
[18]

A Global Geometric Framework for Nonlinear Dimen- sionality Reduction,

J. B. Tenenbaum, V . de Silva, and J. C. Langford, “A Global Geometric Framework for Nonlinear Dimen- sionality Reduction,”Science, vol. 290, no. 5500, Dec. 2000.DOI: 10.1126/science.290.5500.2319

work page doi:10.1126/science.290.5500.2319 2000
[19]

Vector Autoregressive Models with Structural Changes in Regression Coefficients and in Variance- Covariance Matrices,

J. Bai, “Vector Autoregressive Models with Structural Changes in Regression Coefficients and in Variance- Covariance Matrices,”CEMA Working Papers, no. 24, Oct. 2000

2000
[20]

Optimal Transport for structured data with application on graphs,

V . Titouan, N. Courty, R. Tavenard, C. Laetitia, and R. Flamary, “Optimal Transport for structured data with application on graphs,” inProceedings of the 36th International Conference on Machine Learning, PMLR, May 2019

2019
[21]

Inexact graph matching for structural pattern recognition,

H. Bunke and G. Allermann, “Inexact graph matching for structural pattern recognition,”Pattern Recognition Letters, vol. 1, no. 4, May 1983.DOI: 10.1016/0167- 8655(83)90033-8 APPENDIX A. Additional Results for CPD on Synthetic Data Tables IV to VI compare change detection metrics for multiple algorithms on synthetic SBM trajectories, withour methodsin ita...

work page doi:10.1016/0167- 1983

[1] [1]

Anomaly detection in dynamic networks: A survey,

S. Ranshous, S. Shen, D. Koutra, S. Harenberg, C. Faloutsos, and N. F. Samatova, “Anomaly detection in dynamic networks: A survey,”WIREs Computational Statistics, vol. 7, no. 3, 2015.DOI: 10.1002/wics.1347

work page doi:10.1002/wics.1347 2015

[2] [2]

A Survey of Change Point Detection in Dynamic Graphs,

Y . Zhou, S. Gao, D. Guo, X. Wei, J. Rokne, and H. Wang, “A Survey of Change Point Detection in Dynamic Graphs,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 3, Mar. 2025.DOI: 10.1109/TKDE.2024.3523857

work page doi:10.1109/tkde.2024.3523857 2025

[3] [3]

Selective review of offline change point detection methods,

C. Truong, L. Oudre, and N. Vayatis, “Selective review of offline change point detection methods,”Signal Pro- cessing, vol. 167, Feb. 2020.DOI: 10.1016/j.sigpro. 2019.107299

work page doi:10.1016/j.sigpro 2020

[4] [4]

Transfer Graph Neural Networks for Pandemic Forecasting,

G. Panagopoulos, G. Nikolentzos, and M. Vazirgian- nis, “Transfer Graph Neural Networks for Pandemic Forecasting,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 6, May 2021.DOI: 10.1609/aaai.v35i6.16616

work page doi:10.1609/aaai.v35i6.16616 2021

[5] [5]

A pruned dynamic programming algorithm to recover the best segmentations with 1 to K max change-points.,

G. Rigaill, “A pruned dynamic programming algorithm to recover the best segmentations with 1 to K max change-points.,”Journal de la Soci ´et´e Franc ¸aise de Statistique, vol. 156, no. 4, Dec. 2015

2015

[6] [6]

A Cluster Analysis Method for Grouping Means in the Analysis of Variance,

A. J. Scott and M. Knott, “A Cluster Analysis Method for Grouping Means in the Analysis of Variance,” Biometrics, vol. 30, no. 3, 1974.DOI: 10.2307/2529204

work page doi:10.2307/2529204 1974

[7] [7]

Optimal De- tection of Changepoints With a Linear Computational Cost,

R. Killick, P. Fearnhead, and I. A. Eckley, “Optimal De- tection of Changepoints With a Linear Computational Cost,”Journal of the American Statistical Association, vol. 107, no. 500, Dec. 2012.DOI: 10.1080/01621459. 2012.737745 10

work page doi:10.1080/01621459 2012

[8] [8]

Concept Drift and Anomaly Detection in Graph Streams,

D. Zambon, C. Alippi, and L. Livi, “Concept Drift and Anomaly Detection in Graph Streams,”IEEE Trans- actions on Neural Networks and Learning Systems, vol. 29, no. 11, Nov. 2018.DOI: 10 . 1109 / TNNLS . 2018.2804443

arXiv 2018

[9] [9]

R. P. Duin et al.,The Dissimilarity Representation for Pattern Recognition: Foundations and Applications. World scientific, 2005, vol. 64

2005

[10] [10]

T. Chen, Z. Lubberts, A. Athreya, Y . Park, and C. E. Priebe,Euclidean mirrors and first-order changepoints in network time series, Oct. 2024.DOI: 10.48550/arXiv. 2405.11111

work page internal anchor Pith review doi:10.48550/arxiv 2024

[11] [11]

A Comprehensive Survey on Graph Anomaly Detection With Deep Learning,

X. Ma et al., “A Comprehensive Survey on Graph Anomaly Detection With Deep Learning,”IEEE Trans- actions on Knowledge and Data Engineering, vol. 35, no. 12, Dec. 2023.DOI: 10.1109/TKDE.2021.3118815

work page doi:10.1109/tkde.2021.3118815 2023

[12] [12]

Statis- tical inference for Bures–Wasserstein barycenters,

A. Kroshnin, V . Spokoiny, and A. Suvorikova, “Statis- tical inference for Bures–Wasserstein barycenters,”The Annals of Applied Probability, vol. 31, no. 3, Jun. 2021. DOI: 10.1214/20-AAP1618

work page doi:10.1214/20-aap1618 2021

[13] [13]

On the Bures– Wasserstein distance between positive definite matri- ces,

R. Bhatia, T. Jain, and Y . Lim, “On the Bures– Wasserstein distance between positive definite matri- ces,”Expositiones Mathematicae, vol. 37, no. 2, Jun. 2019.DOI: 10.1016/j.exmath.2018.01.002

work page doi:10.1016/j.exmath.2018.01.002 2019

[14] [14]

GOT: An Optimal Transport framework for Graph comparison,

H. Petric Maretic, M. El Gheche, G. Chierchia, and P. Frossard, “GOT: An Optimal Transport framework for Graph comparison,” inAdvances in Neural Information Processing Systems, vol. 32, Curran Associates, Inc., 2019

2019

[15] [15]

Bures-Wasserstein Means of Graphs,

I. Haasler and P. Frossard, “Bures-Wasserstein Means of Graphs,” inProceedings of The 27th International Con- ference on Artificial Intelligence and Statistics, PMLR, Apr. 2024

2024

[16] [16]

At- tribute Fusion in a Latent Process Model for Time Series of Graphs,

M. Tang, Y . Park, N. H. Lee, and C. E. Priebe, “At- tribute Fusion in a Latent Process Model for Time Series of Graphs,”IEEE Transactions on Signal Processing, vol. 61, no. 7, Apr. 2013.DOI: 10 . 1109 / TSP. 2013 . 2243445

2013

[17] [17]

Laplacian Change Point Detection for Dynamic Graphs,

S. Huang, Y . Hitti, G. Rabusseau, and R. Rab- bany, “Laplacian Change Point Detection for Dynamic Graphs,” inProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Virtual Event CA USA: ACM, Aug. 2020. DOI: 10.1145/3394486.3403077

work page doi:10.1145/3394486.3403077 2020

[18] [18]

A Global Geometric Framework for Nonlinear Dimen- sionality Reduction,

J. B. Tenenbaum, V . de Silva, and J. C. Langford, “A Global Geometric Framework for Nonlinear Dimen- sionality Reduction,”Science, vol. 290, no. 5500, Dec. 2000.DOI: 10.1126/science.290.5500.2319

work page doi:10.1126/science.290.5500.2319 2000

[19] [19]

Vector Autoregressive Models with Structural Changes in Regression Coefficients and in Variance- Covariance Matrices,

J. Bai, “Vector Autoregressive Models with Structural Changes in Regression Coefficients and in Variance- Covariance Matrices,”CEMA Working Papers, no. 24, Oct. 2000

2000

[20] [20]

Optimal Transport for structured data with application on graphs,

V . Titouan, N. Courty, R. Tavenard, C. Laetitia, and R. Flamary, “Optimal Transport for structured data with application on graphs,” inProceedings of the 36th International Conference on Machine Learning, PMLR, May 2019

2019

[21] [21]

Inexact graph matching for structural pattern recognition,

H. Bunke and G. Allermann, “Inexact graph matching for structural pattern recognition,”Pattern Recognition Letters, vol. 1, no. 4, May 1983.DOI: 10.1016/0167- 8655(83)90033-8 APPENDIX A. Additional Results for CPD on Synthetic Data Tables IV to VI compare change detection metrics for multiple algorithms on synthetic SBM trajectories, withour methodsin ita...

work page doi:10.1016/0167- 1983