Computer Vision News - December 2016

CVN: On the other hand, there may be students who do not learn enough about the basics of science, math, physics, geometry, etc. and something is getting lost. Paragios: Unfortunately, that is what is happening. Now, the way the papers are evaluated is based on how well they do on benchmarks. The objective becomes to produce the best possible results. If you don’t have a principle based on a mathematically rigorous way to understand what you are doing, then you’re just going to keep spending time on trying. It’s something that has its own value, but in the long run it’s important that every PhD student or researcher knows where the field comes from: deep learning now is hot, but if I look 10 years back, everyone was doing compressed sensing and everyone was saying that compressed sensing was the future. What is happening now, is that compressed sensing is still there, but the impact is not there yet. It’s important that people know what has been done before. Perhaps a wide theoretical view of the problem is better than a constrained view of the visual field by just applying these methods to different problems. CVN: Scientists are trying to teach computers to mimic how our brains work. I know of a 5 years old kid, who could immediately recognize what another dinosaurs looked like from one toy dinosaur he had. A computer needs to see a massive number of dinosaurs until he starts to recognize them with some kind of certainty. Does it mean that the computer is nowhere near what a little kid can comprehend? Paragios: There are two ways of answering your question. Actually, we don’t have a real clue on how the brain works. We have ideas on how decisions are made. We have some ideas on the connectivity. We have some ideas on the computing power, but we don’t really know on a very fine scale what is exactly happening, because we don’t have the tools to visualize how decisions are made. Another way to look at this question is that it really depends on the task. Human intelligence is not only vision, it’s a combination of things . We understand the environment. We use our experience. We use plenty of other sources of information. We have a lot of work that we do in order to actually approach this. I don’t think it’s only a problem of data. There should be something more fundamental that we are not getting yet. That’s why we are really far. CVN: Is it a kind of failure of science that we do not know yet how the brain functions? On what grounds are we sending kids to school where they will be taught things, if we still don’t know how their brain functions? “ There should be something more fundamental that we are not getting yet. That’s why we are really far ” Computer Vision News Guest 5 Guest