On a tangent: elementry geometry diagrams, have straight lines and vertices. It's a graph (discrete).
In the same way, objects in images, if given a minimum possible length (think artificial Planck length given by the encoder/renderer"), can also be represented as graphs.
And matrices are a good way to represent graphs. And GPUs, matrices are used heavily in computer vision.
Is this (images/data <--> matrices <--> ML) a necessary connection? In other words, since graphs are the most general data structures, they are so prevalent in ML and non-optional for multimodal learning.
On a tangent: elementry geometry diagrams, have straight lines and vertices. It's a graph (discrete).
In the same way, objects in images, if given a minimum possible length (think artificial Planck length given by the encoder/renderer"), can also be represented as graphs.
And matrices are a good way to represent graphs. And GPUs, matrices are used heavily in computer vision.
Is this (images/data <--> matrices <--> ML) a necessary connection? In other words, since graphs are the most general data structures, they are so prevalent in ML and non-optional for multimodal learning.