Pierrick Legrand

Prof. Dr. Pierrick Legrand


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications

Verified
Conference Paper
Open access
PDF file
ÉGÉric Grivel
Éric Grivel
PLPierrick Legrand
Pierrick Legrand
Detrended fluctuation analysis (DFA) is a widely used method for estimating the Hurst exponent of time series exhibiting nonstationary trends. In its standard formulation, DFA relies on the least-squares (LS) criterion at several stages of the procedure, including trend removal and Hurst exponent estimation. In practice, the estimation is often sensitive to the practitioner’s choice of local trend sizes, which can significantly affect the estimated Hurst exponent. Moreover, the LS criterion may be poorly suited in the presence of outliers or limited sample sizes. To address these limitations, we propose robust extensions of the DFA. They are based either on a least trimmed squares method or on an iterative weighted LS approach. In the latter case, the weights are defined using Cook distance. This allows all scales to be considered without requiring any a priori selection of local trend sizes. This strategy also improves the stability of the regression step without introducing too many tuning parameters. Illustrations are provided to demonstrate the relevance of the proposed approaches, especially in the presence of outliers.

8 December 2025 in arXiv
Verified
Preprint
Open access
PDF file
YBYann Bourdin
Yann Bourdin
PLPierrick Legrand
Pierrick Legrand
FRFanny Roche
Fanny Roche
This paper investigates the optimization of Truncated Backpropagation Through Time (TBPTT) for training neural networks in digital audio effect modeling, with a focus on dynamic range compression. The study evaluates key TBPTT hyperparameters -- sequence number, batch size, and sequence length -- and their influence on model performance. Using a convolutional-recurrent architecture, we conduct extensive experiments across datasets with and without conditionning by user controls. Results demonstrate that carefully tuning these parameters enhances model accuracy and training stability, while also reducing computational demands. Objective evaluations confirm improved performance with optimized settings, while subjective listening tests indicate that the revised TBPTT configuration maintains high perceptual quality.