9 de mar. de 2012

Reflections on Computer Science, Hopcroft (J E Hopcroft)


Reflections on Computer Science, Hopcroft (J E Hopcroft)
- Highlight on Page 9 | Added on Thursday, November 10, 2011, 01:49 AM

A deeper understanding of the process of knowledge fusion will be achieved in the near future, and this will have a significant impact on computer science. Humans instantaneously comprehend images; when we see a tree, we immediately recognize it as one. Since trees come in an infinite variety of 10 HOPCROFT shapes, it is obvious that our recognition does not depend upon geometrical pattern-matching. Moreover, we recognize trees we have never seen before. What makes a tree a tree is not its shape and texture alone, but something more intrinsic. Our recognition of the tree involves the fusion of many bits of information, some that support the decision for treeness and some that do not. By some as yet unknown technique, humans immediately combine various sensory data to arrive at a correct decision. This combination algorithm must be far from the bottom-up methods of object recognition Annu. Rev. Comput. Sci. 1990.4:1-12. Downloaded from www.annualreviews.org by 189.12.124.179 on 10/30/11. For personal use only. used in computer vision systems today, which identify features such as edges and corners and try to combine these features into more complex patterns. Somehow the human computer possesses top-down information about what is most likely in view. (For example, if one is outdoors, an object is more likely to be a tree than if one is indoors.) The human computer simultaneously processes bottom-up information about what it is seeing and top-down information about what it expects to see; then it fuses these two computations. The fusion probably occurs at many differ ent levels rather than at a single interface. Our recognition processes are rarely fooled; and when they are, the error is usually only momentary. For example, when looking out a train window, we may believe that our train is moving when a train on an adjacent track starts to move. However, since additional information does not jibe with this belief our brain immediately corrects its impression. No computer program yet written has capabilities so advanced, but I foresee developments like this in the near future.