Hi, > What do you estimate are the computational costs of human level > vision? How many nodes and links on the sub-symbolic and symbolic > sides? How many operations per second? How much memory? What training > data will you use? How much? How do you plan to collect it? > > What do you estimate are the computational costs for simpler problems > like reading text, recognizing faces, driving a car, or flying a > drone? Do you plan any such experiments?
On the purely vision side, it seems we can get a big enough DeSTIN hierarchy running on a recent-model GPU card (say, an Nvidia Kepler ...) ... On the symbolic side, I'm less sure, but I'm guessing that if the Atomspace is optimized (as Linas is planning) we could do it on a single contemporary server with, say, 6 or 8 processors and maxed-out RAM..... There are no rigorous calculations for this, though, it will have to be determined via experiment > In one paper you mention SIFT briefly (model of a fovea). But in your > more recent papers there is no mention of it, but rather of optimizing > the DeSTIN learning algorithm to eliminate redundant copies of > features that differ only in translation, scale, and rotation. SIFT > would not be compatible with translation invariance, but (IMHO) would > be a better optimization. Has there been any more work on SIFT? We shifted to the other approach that you like less -- i.e. we have tweaked DeSTIN to make its pattern-recognition intrinsically translation invariant... Scale invariance will come next, after depth and color vision.... The code for porting a pattern from one level of the hierarchy to another has already been written & tested... The person who implemented translation invariance went back to his day job (he was working on DeSTIN temporarily full time, and is now back to very-part-time), and we have a new DeSTIN person, so progress is a bit slow while he gets up to speed... but the stuff will get done... I am unsure why you call SIFT a model of a fovea.. We have discussed giving the DeSTIN hierarchy greater resolution in the center of the grid, which would be a clearer and simpler model of a fovea, I think... -- Ben G ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
