------------------------------------------------------------------ Parallel Algorithm Discovery and Orchestration Astro Teller Manuela Veloso Computer vision and machine learning seem to be two disiplines that are made for each other. Yet dispite a healthy number of different approaches that have been tried, computer vision continues to be one of AI's most frustrating areas and has received only minor support from machine learning. One of the main goals of computer vision must be the solution of the signal to symbol problem. That is, the ability to recognize what object(s) are shown in the image. A variety of architectures have been tried in the past and none have found any real success with natural objects in natural settings. (i.e. most "success" have come from the recognition of geometric objects or objects whose 3-D aspects have been careful recorded) This talk has two purposes. The first is to introduce a new architecture and paradigm for learning how to solve complex tasks. This paradigm is called PADO. PADO's novelty lies in its ability to generate ALGORITHMS that produce useful sub-solutions to difficult problems. The second purpose of the talk is to show how PADO can be applied to a complex problem like the recognition of everyday objects in a natural setting. --------------------------------------------------------------------