Created: January 16, 2021
Modified: January 24, 2022
Modified: January 24, 2022
blur your eyes
This page is from my personal notes, and has not been specifically reviewed for public consumption. It might be incomplete, wrong, outdated, or stupid. Caveat lector.- Jitendra Malik would often tell us that some idea or explanation makes sense if you 'blur your eyes'. This seems counterintuitive, but it's a deep truth.
- Fundamentally it relates to the use of abstraction. Something with too many moving parts is hard to understand; you have to zoom out and focus on larger-scale structures. Blurring your eyes allows you to see the map, not the territory; the forest, not the trees.
- In AI the idea comes up more directly in coarse-to-fine algorithms. These start with a low-resolution image, or in other areas, a small number of clusters or graph or tree nodes or data points, a low-dimensional space, etc., and gradually refine until they reach a final, finer answer. This is doubly advantageous:
- it improves results: the coarser representation removes irrelevant detail, improves statistical strength, forces attention to the big picture, and ensures that it's possible to make consequential moves.
- it makes the algorithm cheaper: working with the coarse representation is cheaper than working with the fine representation.