> > The whole area of machine learning in vision is both huge and balkanised. > Numenta began by looking at things like vision but quickly retreated as NuPIC > was nowhere near ready to take it on. Jeff has concentrated instead on > building a model of a single layer of a cortical region and seeing how far > that would take us, keeping as close as possible to the neuroscience for the > design.
Which I think will eventually solve vision problem, since our brain does it anyways. > > The result is that NuPIC has some visual capabilities, but they are very > limited, in terms of what you're talking about, by the lack of hierarchy in > NuPIC at present. You mean hierarchy in terms of creating several regions and connecting them in a hierarchy? Isn't this one of the key points for HTM? > 1. Unsupervised training of a single layer Restricted Boltzmann Machine to > create a layer of feature detectors. > 2. Stack a few more layers on top, again each is trained unsupervised on the > input from the layer below. > 3. Connect a "label" layer to a joint top associative layer to learn > categories (or objects), use simple backpropagation to tune the RBM stack. That's the idea. Here is how I planned to do this with HTM: 1) Single layer HTM with and input from label 2) Single layer HTM with an input of image (have to solve invariance here, so similar objects yield in the similar columns being activated) 4) Single layer HTM that takes 1 and 2 as combined input, so it would be possible to feed in the label and get the image. > I don't see the logic in looking at NuPIC at the top of your pipeline, to be > honest. You're already worried about the quality of the feature detection > coming in from OpenCV, so I'd suggest considering either RBM's or HTM's at > the feature detection and categorisation level first. HTM in general is something I feel comfortable working with. My background is more in neuroscience and psychology. I lack deep background in computer science and machine learning. HTM just feels natural. I know this is not right, but I'm trying to solve all problems with layers and hierarchies of HTM. I was thinking yesterday that an image categorization problem can be described with a complex hierarchy that takes into account many factors of an image. For us, humans, context simplifies object categorization. I believe if image is analyzed contextually (textures, color scheme, shapes, colors) it would yield in better recognition, as image becomes more meaningful and informative for an HTM network. _______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
