KL divergence between the trained posterior distribution obtained by infinitesimal step size gradient descent and a Gaussian prior.

However,GNNs have notconsidered different types ofuncertainties associated with class probabilities to minimize risk of increasing misclassification under uncertainty in real life. In this work, we propose a multi-source uncertainty framework using a GNN that reflects various types of predictive uncertainties in both deep learning and belief/evidence theory domains fornodeclassification predictions.

One of the central questions of perception is how organisms reliably estimate hidden or abstract quantities ofinterest usingnoisyandambiguous sensory information. Almost equally important is representing the reliability of these estimates, especially in complex environments and situations ofrisk,where theuncertainty associated withachoice mayradically change theoptimal course of action.

Many (if not all) real-world applications require efficient handling of uncertainty and a compact representation of a wide variety of knowledge. Indeed, complex concepts and relationships that typically comprise expert knowledge may be difficult to express in graphical models but can be represented compactly using classical logic.