Page 49 - Artificial Intelligence in the Age of Neural Networks and Brain Computing
P. 49

36     CHAPTER 2 Mind, Brain, Autonomous Agents, and Mental Disorders








































                         FIGURE 2.3
                         A model macrocircuit of an emerging unified theory of visual intelligence. Its processing
                         stages begin at the retina and end in the prefrontal cortex, and include both the What and
                         the Where cortical processing streams. The bottom-up, horizontal, and top-down
                         interactions between model processing stages are needed to overcome the computational
                         weaknesses that each processing stage would experience if it acted alone, due to
                         complementary computing.



                         4. THE WHAT AND WHERE CORTICAL STREAMS ARE
                            COMPLEMENTARY

                         The category learning, attention, recognition, and prediction circuits of the ventral,
                         or What, cortical processing stream for perception and cognition [9,10] are compu-
                         tationally complementary to those of the dorsal, or Where and How, cortical
                         processing steam for spatial representation and action [9e11]. One reason for
                         this What-Where complementarity is that the What stream learns object recognition
                         categories that are substantially invariant under changes in an object’s view, size, and
                         position. These invariant object categories enable our brains to recognize valued
                         objects without experiencing a combinatorial explosion. They cannot, however, locate
   44   45   46   47   48   49   50   51   52   53   54