MIT Researchers Are Creating A Neural Net For Smartphones
Speech recognition, face recognition, and many other artificial intelligence (AI) applications have been improving recently, and artificial neural networks (ANNs) are behind those improvements. However, ANNs are so large and use so much power there’s no way they can be housed in today’s smartphones. That means that any apps that require ANNs must have the…
This story was produced through MarketScale. See how Energy teams put it to work with Customer Stories & Case Studies.
Speech recognition, face recognition, and many other artificial intelligence (AI) applications have been improving recently, and artificial neural networks (ANNs) are behind those improvements. However, ANNs are so large and use so much power there’s no way they can be housed in today’s smartphones. That means that any apps that require ANNs must have the ANNs hosted on an external server, with the data downloaded to that server so the ANN can process the data and then send the new data back to the smartphone.
The problem is that processors store memory in one part of the chip and engage in processing on another part of the chip. So, data must be shuttled back and forth for any computation. The more computations, the more energy used. In response, processors have increased in power as well as energy usage and heat production. ANNs do so many computations that no smartphone could possibly host a processor capable of running one.
All of that may soon change. Researchers at MIT have developed a new kind of chip that may soon bring ANNs directly into your smartphones. As Campus Technology reports, “The new chip improves the speed of neural-network computation by three to seven times and reduces energy consumption by 94 to 95 percent.”
The way the problem has been solved—assuming the solution scales to much larger ANNs than the 16-node ANNs in the prototype—is by simplifying the ANN computations to a dot product and implementing the dot product functionality inside the memory itself. That means data does not have to be transferred back and forth for computations to take place.
“In the new chip, input values are converted to voltage and multiplied by weight and then added together. The voltage is only then converted back to data for storage and further processing. The process allows the new chip to figure the dot products for multiple nodes in one step, eliminating the need to shovel data back and forth repeatedly.”
As advancements continue in the development of these new chips, smartphones will be able to house their own ANNs and thus be true AI machines in their own rights. In this sense, we may be on the cusp of transforming your smartphone into a truly smart phone.
