The latest member of the VIEWS team, Simon Polichinel von der Maase’s research spans the field of conflict studies, data science, and machine learning. He works on conflict forecasting with a focus on Gaussian processes and cutting-edge deep learning architectures – all in a generative and probabilistic setting.

Given this expertise, the VIEWS prediction models will soon be complemented with neural networks. First of which is ConflictNet: a recurrent U-net that leverages convolutional layers to capture spatial patterns of conflict and connect them through time with a gated recurrent network structure. Currently, inputs include conflict history for state-based conflict, non-state conflict, and one-sided violence. Outputs include both the predicted magnitudes of fatalities and the probability of all input features – represented as approximate Bayesian posterior distributions to account for inherent uncertainty. Importantly, the architecture is both flexible and scalable which enables easy expansion of input and output features, paving the way for forecasting derived conflict effects.