This paper revisit the methodology of system identification and shows how new paradigms from machine learning can be used to improve the model identification performance in the case of non-linear systems observed with noisy and unbalanced dataset. We prove that using importance sampling schemes in system identification can provide significant performance boost on a wide variety of systems, in particular when some of the system dynamic is only exhibited by relatively rare events. The performance of the approaches is evaluated on a real and simulated drone and two standard datasets from real robotic systems. Our approach consistently outperforms baseline approaches on these datasets, all the more when the datasets are noisy and unbalanced.

Filling large data-gaps in Micro-Meteorological data has mostly been done using interpolation techniques based on a marginal distribution sampling. Those methods work well but need a large horizon of the previous events to achieve good results since they do not model the system but only rely on previously encountered iterations. In this paper, we propose to use multi-head deep attention networks to fill gaps in Micro-Meteorological Data. This methodology couples large-scale information extraction with modeling capabilities that cannot be achieved by interpolation-like techniques. Unlike Bidirectional RNNs, our architecture is not recurrent, it is simple to tune and our data efficiency is higher. We apply our architecture to real-life data and clearly show its applicability in agriculture, furthermore, we show that it could be used to solve related problems such as filling gaps in cyclic-multivariate-time-series.

No abstract

No abstract