3IA PhD/Postdoc Seminar #15

Neural information coding for efficient spike-based image denoising

Abstract: 
In recent years, Deep Convolutional Neural Networks (DCNNs) have outreached the performance of classical algorithms for image restoration tasks. However most of these methods are not suited for computational efficiency and are therefore too expensive to be executed on embedded and mobile devices. In this work we investigate Spiking Neural Networks (SNNs) for Gaussian denoising, with the goal of approaching the performance of conventional DCNN while reducing the computational load. We propose a formal analysis of the information conversion processing carried out by the Leaky Integrate and Fire (LIF) neurons and we compare its performance with the classical rate-coding mechanism. The neural coding schemes are then evaluated through experiments in terms of denoising performance and computation efficiency for a state-of-the-art deep convolutional neural network. Our results show that SNNs with LIF neurons can provide competitive denoising performance but at a reduced computational cost. .

Environmental factors affecting wildfire-burned areas in southeastern France

Abstract:
Forest fires burn an average of about 440 000 ha each year in southern Europe. These fires cause numerous casualties and deaths and destroy houses and other infrastructure. In this study, we investigated the spatiotemporal evolution in the burned area over a 50-year period (1970–2019) and its interactions with topography and vegetation type in southeastern France by exploiting the geographic information system databases. Data were analyzed for two 25-year periods (1970–1994 and 1995–2019), since a new fire suppression policy was put into place after 1994, which focused on rapid extinction of fires in their early phase. In the last 25 years, the burned area decreased sharply, and the geographic distribution of fires also changed, especially in regions where large fires occur (Var administrative division). Elsewhere, even though forest fires remain frequent, the total extent of the burned area decreased substantially. Fire hotspots appear closer to built-up areas in the west, are randomly distributed in the east, and almost completely disappear in the central region of the study area where there is a history of large fires. Slope orientation presents an increasingly important role in the second period; south-facing slopes are preferred the most by fire, and north-facing slopes are preferentially avoided. The greatest proportion of the burned area is strongly associated with the location of sclerophyllous vegetation clusters which expand in the area over time. 

Open discussion on the two contributions