Memory networks are a recently introduced neural network architecture that can read multiple times from an external memory before producing an output. I will discuss how they are have been used, mostly for tasks in natural language processing, and then briefly speculate how they may be useful for applications in shape analysis and 3D geometry. In particular, because they can operate on arbitrary, unstructured data, they are well suited for applications where one wants to input a set of points, or perhaps some multiscale description of a set of points, and output a fixed length vector.
Back to Shape Analysis and Learning by Geometry and Machine
Copyright. All Rights Reserved.