Random matrix theory improved Fréchet mean of symmetric positive definite matrices

Authors: Florent Bouchard, Ammar Mian, Malik Tiomoko, Guillaume Ginolhac, Frederic Pascal

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental evaluation, involving both synthetic and real-world EEG and hyperspectral datasets, shows that we largely outperform stateof-the-art methods.
Researcher Affiliation Collaboration 1Universit e Paris Saclay, CNRS, Centrale Sup elec, L2S 2Universit e Savoie Mont Blanc, LISTIC 3Huawei Paris Research Center.
Pseudocode Yes Algorithm 1 Covariance based on RMT corrected distance, Algorithm 2 RMT corrected Fr echet mean on S++ p, Algorithm 3 Nearest Centroid classifier based on RMT, Algorithm 4 K-Means clustering based on RMT
Open Source Code Yes To ensure reproducibility, the code for the experiments discussed is accessible at https://github.com/ Ammar Mian/icml-rmt-2024.
Open Datasets Yes motor imagery datasets accessible via the MOABB platform (Aristimunha et al., 2023).
Dataset Splits No The paper mentions a 'training set' in the context of the Nearest Centroid algorithm, but it does not provide specific train/validation/test split percentages, sample counts, or explicit instructions for how the datasets were partitioned for the empirical evaluations.
Hardware Specification No The development and evaluation of these methods were conducted in Python. Specifically, SCM and LW implementations were sourced from the scikit-learn library (Pedregosa et al., 2018), while LW-NL comes from scikit-RMT3. The conventional Fr echet means, standard Nearest Centroid and K-Means algorithms were taken from the py Riemann library (Barachant et al., 2023).
Software Dependencies No The development and evaluation of these methods were conducted in Python. Specifically, SCM and LW implementations were sourced from the scikit-learn library (Pedregosa et al., 2018), while LW-NL comes from scikit-RMT3. The conventional Fr echet means, standard Nearest Centroid and K-Means algorithms were taken from the py Riemann library (Barachant et al., 2023).
Experiment Setup Yes Parameters are p = 64, K = 10 on the left and n = 128 on the right. (from Figure 1 caption under Simulations) and 'The K-means algorithm, capped at 100 iterations with early stopping at a 10^-4 tolerance'.