Saw this live. On my 14th birthday. At home. No school for some reason. Bad jokes showed up next morning in homeroom. Pre-internet. Good teacher used this as a teachable moment as we were learning about Crispus Attix. “You won’t remember any names but Christy McCaulaffe just as we don’t remember any names other than Crispus Attix”. I argued with her and rattled off all 7 names. But… in the end… she was right. Good teacher.

Saw this live.
On
[read full article]
 

Well, if you want to understand my way of thinking, he nails it. He’s got the rare blend of knowledge, optimism and skepticism that I consider good scientific traits . I think to understand human intelligence, you have to understand intelligence first. Give it a range. Look around at what lives and what are their levels of intelligence and what makes it so. So, one example is learning reflexes. How to have AI habituate? A danger of habituation is making mistakes but I think trying to go straight to human AI without making the mistakes of other intelligent creatures loses on some of the mechanisms that makes ours work.

Well, if you want [read full article]

 

Ok. Since mathematical morphology is a field I’m starting to ‘get’, working within its framework is helpful, as it allows a precise mapping from anything mathematical to a hierarchical graph that can be displayed on a 2D computer screen. So, if I’m understanding it right, CAT(0) spaces are a Hadamard manifold. So, that means EVERY two points can be connected by a minimizing geodesic. (learned that just now in: Riemannian mathematical morphology, J Angulo, S Velasco-Forero – Pattern Recognition Letters, 2014) But I’m learning that success depends upon having a range. You have to chose a subset – a Infimum and supremum — IF I’m understanding the formula right… and knowing a bit about restrictions inherent to displaying images. This will probably give a better clue but I haven’t read it yet. https://arxiv.org/abs/1401.5053

Ok. Since mathematical morphology
[read full article]
 

It is the generation and accuracy of the ground truth corpus and annotation that has a lot of my attention. Once you have ground truth, you can erode and dilate until you find a match, or use heuristics or other segmentation methods. But it’s the “what you are attempting to conform to?” which grabs my fascination, not just with computer vision but with concepts, metaphors, any kinds of matching.

It is the generation [read full article]