Non-parametric Online Change Point Detection on Riemannian Manifolds
Authors: Xiuheng Wang, Ricardo Augusto Borsoi, Cédric Richard
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results with both synthetic and real data illustrate the performance of the proposed method. |
| Researcher Affiliation | Academia | 1Universit e Cˆote d Azur, CNRS, OCA, France 2Universit e de Lorraine, CNRS, CRAN, Vandoeuvre-l es-Nancy, France. |
| Pseudocode | Yes | Algorithm 1 Online CPD on Riemannian manifolds Input: {xt}, step sizes λ, Λ, threshold ξ. Initialization: mλ,0 = mΛ,0 = x0. for t = 1, 2, 3, . . . do Update the generalized Karcher mean estimates mλ,t and mΛ,t using (4) and (5); Compute the test statistic gt = d M(mλ,t, mΛ,t); if gt > ξ then Flag t as a change point; end if end for |
| Open Source Code | Yes | Open-source code to reproduce the results is publicly available at https://github.com/ xiuheng-wang/CPD_manifold_release. |
| Open Datasets | Yes | TIMIT database (Garofolo, 1993), QUTNOISE database (Dean et al., 2010), HDM05 motion capture database (M uller et al., 2007) |
| Dataset Splits | No | The paper does not provide explicit details about train/validation/test splits (e.g., percentages or sample counts). While it mentions 'training data' in the context of adaptive threshold selection, it does not specify how the datasets for the main experiments were split for training, validation, and testing. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory, or cloud instance types) used for conducting the experiments. |
| Software Dependencies | No | Our method was implemented in Python using Pymanopt (Townsend et al., 2016). |
| Experiment Setup | Yes | The step sizes of our method were set as λ = 0.01 and Λ = 0.02. ... In Scan B, the number of reference blocks was set to 3. NEWMA was implemented with Random Fourier features using the Gaussian kernel. The window size of Scan-B and NEWMA were both set to 50. The reference and test window lengths of NODE were both set to 64. |