Page 173 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

1. Introduction and Overview 163

But the purpose of this paper is not to relitigate the past. Section 2 will explain a
few of the key concepts of deep learning, starting from the recent industry interest
and working back to what underlies it. The methods which have become widely used
in the computer industry since 2011 validate the idea of trying to learn fundamental
principles from the brain, but they still fall far short of fully exploiting the funda-
mental design capabilities which can be found even in the brain of the mouse. We
now have a solid mathematical path for how to build such a high level of intelligence
[5], including many powerful methods which have been applied in engineering but
not yet widely disseminated in computer science; Section 3 will discuss a few
examples, with pointers to where you can learn the all-important technical details.

1.3 NEED AND OPPORTUNITY FOR A DEEP LEARNING REVOLUTION
IN NEUROSCIENCE

The deep learning cultural revolution in computer science was a great advance, but
in my view, we still need another cultural revolution in the field of computational
neuroscience, which has yet to assimilate new findings about how intelligence works
at a systems level in the brain, despite a series of NSF efforts to build bridges be-
tween the relevant disciplines [1,2,6], and great efforts by systems neuroscientists
like Karl Pribram and Walter Freeman [7] to broaden the paradigm. Section 4
will review recent research opening the door to a whole new paradigm here [8,9],
and some of the important previous work leading up to it.
Fig. 8.1 illustrates the research goals and strategy of the cross-cutting NSF initia-
tive in cognitive optimization and prediction which I proposed and led in 2008, and
did my best to continue funding as part of my core program in the Engineering

FIGURE 8.1
Vision of the NSF COPN research initiative of 2008 [6].