- author = "Astro Teller and Manuela Veloso",
- institution = "Department of Computer Science, Carnegie Mellon University",
- address = "Pittsburgh, PA, USA",
- title = "PADO: Learning Tree Structured Algorithms for Orchestration into an Object Recognition System",
- number = "CMU-CS-95-101",
- year = "1995",
-
URL = "
http://www.cs.cmu.edu/afs/cs/usr/astro/public/papers/PADO-Tech-Report.ps",
- keywords = "genetic algorithms, genetic programming, memory",
- size = "2.54 Mbytes",
- abstract = "Around the world there are innumerable databases of information. The quantity of information available has created a high demand for automatic methods for searching these databases and extracting specific kinds of information. Unfortunately, the information in these databases increasingly contains signals that have no corresponding classification symbols. Examples include databases of images, sounds, etc. A few systems have been written to help solve these search and retrieve issues. But we can not write a new system for every kind of signal we want to recognize and extract. Some work has been done on automating (i.e. learning) the task of identifying desired signal elements. It would be useful to automate (learn) not just a part of the classification function, but the entire signal identification program. It would be helpful if we could use the same learning architecture to automatically create these programs for distinguishing many different classes of the same signal type. It would be better still if we could use the same learning architecture to create these programs even for signal types as different as images and sound waves. We introduce PADO (Parallel Architecture Discovery and Orchestration), a learning architecture designed to deliver this. PADO has at its core a variant of genetic programming (GP) that extends the paradigm to explore the space of algorithms. PADO learns the entire classification algorithm for an arbitrary signal type with arbitrary signal class distinctions. This architecture has been designed specifically for signal understanding and classification. The architecture of PADO and its achievements on the recovery of visual and acoustic signal classes from test databases are the subjects of this article. Keywords: Machine Learning, Signal Understanding, Data Mining, Genetic Programming, Algorithm Evolution",
-
notes = "You can get these papers by anonymous ftp to any CMU
machine. (e.g. GS61.SP.CS.CMU.EDU (128.2.203.143) or
J.GP.CS.CMU.EDU (128.2.250.198))
then cd to /afs/cs/usr/astro/public/papers/
Since several come from the Mac, they won't work in GhostView, but they should print fine.
Our printer barfed at page 10! This appears to be very close to \cite{teller:1995:PADO}",
