Team leader --> Dominique Pastor
Communication manager --> Pierre Tandeo
"Ocean monitoring using underwater acoustics and spaceborne remote sensing":
The TOMS team has strong expertise both in ocean acoustics and in ocean remote sensing from space. These two themes will be gathered to develop new ocean monitoring methods. In particular, spaceborne remote sensing has an excellent spatial resolution but a poor time resolution. On the other hand, acoustic monitoring has an excellent time resolution but a poor spatial resolution. It is thus promising to merge spaceborne remote sensing with acoustic to infer the best oceanographic information. The potential applications range from physical oceanography (e.g. spatio-temporal tracking of geophysical fields) to marine biology (e.g. marine mammal behavior, benthic ecology). Collaboration with marine science institutes (e.g. Ifremer, IUEM) will be developed.
"Dynamical systems in geophysics":
This TOMS working group is working on different aspects of the dynamical systems. First aspect is the statistical emulation of the dynamics using large datasets of historical observations of the system. In geophysics, we talk about analog forecasting (instead of model forecasting) where a statistical combination of nearest neighbors will provide estimates of the future state. Then, we combine these forecasting procedure to data assimilation procedures. Second point of interest is the evaluation of errors in dynamical systems. Indeed, most of the time, the differential equations are approximations of the true dynamics of the systems. A key point is to statistically evaluate their uncertainties and take them into account in data assimialtion procedures. Last point concerns the multiscale aspects. Most of the time, observation resolutions of the system are high in space or in time. These observations give precious details on the true underlying model. Statistical methods can be usefull to link different scales and take into account these nonlinear and high resolution phenomena.
"From signal processing to conservation of the biodiversity":
Marine bioacoustic is at the cross road of different methodologies including signal processing, oceanography and biology. Passive acoustic monitoring could be dedicated to promote a better understanding of highly exploited and endangered marine species such as baleen whales. Marine Mammals produce a wild variety of signals from short impulsive echolocation clics to long tonal calls. The goal of this working group is to join people from two communities, signal processing and biology, to bring and exchange ideas, concepts, methods and needs in the ultimate perpective to improve passive acoustic monitoring tools.
"Machine learning and novelty detection":
Machine learning is a hot topic, especially because of recent results in deep learning and advances in neural networks. However, these results also raise some questions that we hereafter encompass under the general term of novelty detection. Basically, as detailed in “A review of novelty detection”, by Pimentel et al., published in Signal Processing 99 (2014) 215–249, novelty detection amounts to detecting data that may differ, in some extent to define, from those used during the training. Novelty detection involves anomaly and outlier detection. It extends to the detection of data representing new knowledge to integrate dynamically or incrementally in the training, which may be crucial in applications. Another issue is performance prediction. In machine learning, performance is usually measured on test data. Such performance measurements are thus valid for the available training and the test database. Can we then predict the performance of the system when this one is confronted to data that may contain outliers, anomalies or data representing some new kind of knowledge to memorize? In TOMS, new results and paradigms have been proposed, studied and developed, in nonparametric and robust statistical signal processing and spectral analysis. The question is thus whether these advances can bring solutions to issues such as those mentioned above. In this respect, the working group “Novelty detection & dynamical machine learning” is aimed at gathering researchers interested by this topic, via seminars where specialists exterior to TOMS as well TOMS members working in this field will present and confront their results.