Graph Neural Networks Do Not Always Oversmooth

However, GNNs suffer from the problem of over-smoothing, the property that features of all nodes exponentially converge to the same vector over layers, prohibiting the design of deep GNNs.