Document-level relation extraction with graph neural networks faces a fundamental graph construction gap between training and inference - the golden graph structure only available during training, which causes that most methods adopt heuristic or syntactic rules to construct a prior graph as a pseudo proxy. In this paper, we propose $\textbf{ConstGCN}$, a novel graph convolutional network which performs knowledge-based information propagation between entities along with all specific relation spaces without any prior graph construction. Specifically, it updates the entity representation by aggregating information from all other entities along with each relation space, thus modeling the relation-aware spatial information. To control the information flow passing through the indeterminate relation spaces, we propose to constrain the propagation using transmitting scores learned from the Noise Contrastive Estimation between fact triples. Experimental results show that our method outperforms the previous state-of-the-art (SOTA) approaches on the DocRE dataset.

It is used to disguise a person's identity, cover explicit areas of an image or to render vehicle number plates unreadable. But deliberate pixilation of photographs could soon be rendered useless by artificial intelligence that can peer through the blurring to see what is hidden beneath. Software engineers have used machine learning to teach a piece of software to adapt image recognition techniques to recognise objects, faces and words in obscured images. Artificial intelligence could be used to defeat attempts to protect people's identity (stock image) or hide certain information in videos and photographs posted online. The software could mean that people who appear on Google Street View, for example, could be identified despite attempts by the search company to hide their identity with image blurring. It is a bizarre disappearing act that only the most affluent seem to be able to afford.