Applications

Pointer network architecture. We combine the embedding of each output ID with the embedding of entry points to produce softmax indices for the members of the input point set (author’s image).
The pointer grid consists of a few learnable weights followed by a program symbol in the input vocabulary (an image created by the author https://www.mathcha.io/).
The black arrow with the arrows is our “rubber band” and the blue polygon represents a convex body formed by fitting this “rubber band” to our set of points (source: Wikipedia).
An example of a convex frame created by our pointer network-based convex frame solution n = 50 (author’s picture).
An example of a convex body produced by our pointer network when n = 10. Errors for fewer points lead to larger deviations (author’s figure).

LEAVE A REPLY

Please enter your comment!
Please enter your name here