David Marr's book Vision

Date:
1979-1980
Reference:
PP/CRI/H/6/18
Catalogue details

Licence: In copyright

Credit: David Marr's book <i>Vision</i>. Source: Wellcome Collection.

    1/20
    January 7, 1980 Dr. David Marr Department of Psychology M. I. T. Cambridge, Massachusetts 02139 Dear David, So pleased to get your long and thoughtful letter. Just a few points — the rest can wait till we raeet. I do agree with you about the granule celllidea (which impressed me very much when I first heard it) and about the zero-crossing idea, tl am not yet quite as enthusiastic about the Y-cell ideas). The problem is that they are difficult to test directly at this stage. This is alESEäy a case where a little faith is needed. I was most interested to hear about Richter and Ullman's theory of the retina. I am especially curious about the role they assign to amacine cells, and how many types of them they think there are. Do ask them to send me a preprint when they have one. I am not quite happy with your preliminary remarks about explicit representation. I have an uncomfor&feble feeling that there is more to it than that. And when it comes to a specific case, such as color, I'm still not clear what to expect. But this is too difficult a subject to discuss here. I do admit to nagging you about addressing because, at bottom, I sense that you are evading an issue which I feel may yield important general conclusions. Without going over the whole topic, let me ask one question: is it true that there is a general theoram which says that an algorithm which can be implemented on a computer with an addresâMîèe memory can always be implemented on a computer with a (finite) associate-net type memory? Even if there is a theoram to this effect (which I doubt) I suspect such an implementation migfot be very clumsy. My point is that there may be whole classes of 0 algorithms which cannot be conveniently implemented on associative- net type memories and that this hardly involves specific proposals as to the details of the net. • ^ /