pollack@giza.cis.ohio-state.edu (Jordan B Pollack) (03/29/91)
>>The fractal compression is done by Barnesly (sp?) from Georgia Tech, I >>think. As far as I can tell, it is a money-making hoax. He won't >>disclose his algorithms unless you pay him lots of money, and I have >>yet to see an example of an actual compressed image (He has a few >>images he likes to show, and they *are* generated by ridiculously >>small quantities of data. But he doesn't show the original.... i >>wonder why...) A fractal is like a large computable number which happens to have a surprisingly small program (like the set n!). These numbers are vanishingly rare, but could be very useful. If we can hook representations up to such numbers, we can get very large "memories". This is the train of thought being publically followed by Kym. Further, a fractal is like a grammar: a parsimeous description of a very large set. "Inverting" such a set (finding the underlying rule) is thus equivalent to the language learnability problem, and since very little progress has been made on formal language acquisition since (Gold 1967) formalized the question, there is very good reason to be suspicious of Barnsley's work, esp when the learning algorithm is not fully disclosed for commercial advantage. On the other hand, the idea of using fractal inversion for language learning is quite appealing to break that logjam! Snake oil comes in all fields, and while his rude capitalist behavior impacts our opinion of Barnsley the person, his discovery is still extremely beautiful. Fractal compression, if it works, is better as a theory of lossy/overgeneralizing human reconstructive memory for than as a technology for image compression. Personally, I think the issueance of a patent should result in the public retraction of incomplete scientific papers, such as Barnsley's and Karmarkar's, but this is an issue for alt.science.ethics... -- Jordan Pollack Assistant Professor CIS Dept/OSU Laboratory for AI Research 2036 Neil Ave Email: pollack@cis.ohio-state.edu Columbus, OH 43210 Fax/Phone: (614) 292-4890