Building Maze Solutions with Computational Dreaming
By Lori Cameron
 

abstract painting of people in open field with lightbulb

Computational dreaming (CD) is inspired by the massively parallel structure and dreaming process of the human brain. CD examines previously observed input data during a “dream phase” executed while normal inputs are shut off when a device is not in use.

“Combining the virtually unbounded parallelism in the human brain with the need to sleep, we conclude that dreaming is a phase of exploration and optimization during which resources operate in nearly perfect parallelism, and are optimized and subset for real-time, awake operation,” write Scott M. Jackson and JoAnn M. Paul of Virginia Tech in their article “Building Maze Solutions with Computational Dreaming” (login may be required for full text) published in the July/August 2017 issue of IEEE Micro.

The authors demonstrate that CD can develop a suitable model from scratch while dreaming and select it for use while awake. CD uses a parallel multiple-instruction architecture during the dream phase and an optimized architecture during the awake phase.

They then developed a CD simulator that solved 15 percent of mazes (ranging from small and simple to large and complex) compared with 2.2 percent solved by random model selection.

Significantly, approximately 50 percent of successful solutions generated by the dream phase of computation, with no prior knowledge, match optimal human-generated solution strategies.

 


 

About Lori Cameron

Lori Cameron is a Senior Writer for the IEEE Computer Society and currently writes regular features for Computer magazine, Computing Edge, and the Computing Now and Magazine Roundup websites. Contact her at l.cameron@computer.org. Follow her on LinkedIn.