RINA - Démonstrateur d’utilisation de l’intelligence artificielle pour la gestion opérationnelle des risques naturels gravitaires géologiques | FEREC 2020

49 A.C Turkmen 2019 [11] "Temporal point processes (TPP) are probabilistic models of such data, namely discrete event sets in continuous time. They have been extended widely to describe patterns through which events (points) interact, and to model side information available in the form of features (marks)." " In this paper, we propose a novel model, FastPoint, for efficient learning and approximate inference in multivariate TPP. We combine the expressiveness of RNNs to model mutual excita- tion (between marks), with well-studied Hawkes processes to capture local (within marks) temporal relationships." " In Recurrent Marked TPP (RMTPP), Du et al. [6] propose to model a multivariate point process via an approximation to the conditional intensity function. This is achieved by an RNN, in their experiments an LSTM [12]. Effectively, the LSTM embeds the event history H t = ( t i , y i ) | t i < t to a vector, on which the conditional intensity function and the conditional distribu- tion of the mark of the next point are calculated. Concretely, they take the conditional intensity λ ∗ ( t ) = exp( v T h j + β ( t − t j ) + b ) where β , b are scalar parameters, v is a vector parameter of appropriate dimension. h j is the output of the LSTM for point t j " Spiking Neural Network (SNN) [10] [7] [4] + See website http://www.xavierdupre.fr/app/ensae_teaching_cs/helpsphinx/ml2a/ td2a_mlplus_snn.html A toolbox is available https://snntoolbox.readthedocs.io/en/latest/guide/intro.html Other Papers to overview