PHAEACO: A cognitive architecture inspired by Bongard’s problems

Phaeaco is a cognitive architecture for visual pattern recognition that starts at the ground level of receiving pixels as input, and works its way through creating abstract representations of geometric figures formed by those pixels. Phaeaco can tell how similar such figures are by using a psychologically plausible metric to compute a difference value among representations, and use that value to group figures together, if possible. Groups of figures are represented by statistical attributes (average, standard deviation, and other statistics), and serve as the basis for a formed and thereafter learned concept (e.g., triangle), stored in long-term memory. Phaeaco focuses on the Bongard problems, a set of puzzles in visual categorization, and applies its cognitive principles in its efforts to solve them, faring nearly as well as humans in the puzzles it manages to solve.