Technology developed by researchers at the University of Waterloo has gotten artificial intelligence (AI) another step closer toward relinquishment from internet and cloud computing. Software produced for AI deep learning is compact enough to fit on mobile computer chips that can be used in a broad array of technologies from smartphones to industrial machinery. These breakthroughs would enable the independent operation of devices from the internet, while utilizing AI that performs at levels equivalent to tethered and neural networks.
Waterloo University researchers feel the breakthrough’s potential is astronomical, and could help ease people’s struggles of getting deep learning AI operational in many fields. Stand-alone deep learning AI can prompt significantly lower data processing, transmission costs, greater privacy, along with utilization in areas where existing technology is currently impractical due to factors like expenses. Deep learning AI emulates the human brain by processing data through several artificial neuron layers, which typically require considerable computational power, memory, and energy in order to function.
Researchers turned to evolutionary forces in nature to increase AI efficiency by inserting the software into a virtual environment, before continually (yet progressively) cutting the technology off of resources. The response from the deep learning AI was an ample adaptation to continue functionality whenever computational power and memory were withdrawn. These networks self-evolve themselves over generations, over which they make themselves smaller in order to better sustain these environmental conditions.
Some results that were recently presented at the International Conference on Computer Vision in Venice, Italy, attained a 200-fold reduction in deep learning AI software size that was used to particular tasks pertaining to object recognition. Upon being applied to a chip embedded in a smartphone, AI with this degree of compaction ably ran its speech-activated virtual assistant, along with other intelligent qualities.
This greatly reduces data usage, while eliminating the need for internet services to operate. In addition, other potential applications ranging from low-cost drones and smart grids to surveillance cameras and manufacturing plants also endured issues regarding streaming sensitive or proprietary data to the cloud that continue to persist.
Filed Under: M2M (machine to machine)
