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

48 APPENDIX B. BIBLIOGRAPHY ABOUT POINT PROCESS RELATED TO DEEP NEURAL NETWOR The inhomogeneous Poisson process extends the homogeneous version by relaxing the condition that the distribution is fixed over time. A particular type of inhomogenous Poisson process, the Hawkes process, aims to model self-excitation and so has a distribution of points that depends on the history Bacry et al. (2015); Laub et al. (2015) " "Recurrent Marked Temporal Point Process (RMTPP) Du et al. (2016) was among the first models to employ an recurrent neural network (RNN) to encode the event history and generate parameters defining the intensity function. RMTPP uses an exponential intensity function, a choice which has also been adopted by (Upadhyay et al., 2018). Other methods have employed piecewise constant functions Li et al. (2018); Huang et al. (2019)." Paper S. Xiao 2017 [12] " Our model interprets the conditional intensity function of a point process as a nonlinear mapping, which is synergetically established by a composite neural network with two RNNs as its building blocks. See Figure B.3 ... We first make an observation that many conditional intensity functions can be viewed as an in- tegration of two effects: i) spontaneous background component inherently affected by the internal (time-varying) attributes of the individual and the event type; ii) effects from history events " Figure B.3: RNN model from paper [12] Paper Guen 2020 [6] No point process modeling .No relevant