Robert Lockhart is a New York University film student entering the field of robotics. He hopes, one day, to apply NKS to what AI researchers call “the hard problem,” which is consciousness. Until then, he works at the Apple Store on 5th Avenue and writes occasionally for Robot Magazine. He is preparing himself to enter graduate school for robotics next year.
Project: Image Recognition with 2D Cellular Automata
For this project, 5-neighbor 2D CAs will be used to answer simple existence questions about texture images. “Posing the question” will consist of applying the cellular automaton rules. However, no assumptions will be made about the field surrounding the image, so the input dimensions will each decrease by two on every iteration. With input dimensions kept odd and equal, there will be one cell remaining at the end. This will constitute the “answer” to the question. These answers will then be intersected to find sets of rules that reliably distinguish between textures.
This method is inspired by Hubel & Weisel’s 1959 electrophysiology experiment, which elucidated the visual system and gave rise to a theory of how information flows through the early part of the visual stream. In a way the intent is to reconstruct this experiment within the computational universe.
First, these cellular automaton rules will be applied to very simple input, in order to get a sense of the rule space and which rules could be potentially “useful.”
Later, there will be an attempt to develop an algorithm consisting of a rule or series of rules that can reliably distinguish between satellite images of different regions. This will involve much automatic filtering.
2D Cellular Automata for Texture Recognition
2D Cellular Automata for Feature Detection
Favorite Radius 3/2 Rule
I like this rule because there is both order and disorder. There is obvious repetitive structure, but no obvious pattern to what the next repetitive structure will be.