Artificial intelligence, neural networks, and deep learning continue to gain ground among the giants of the web, but beyond their impressive advances, these technologies need to reach the public in a simple and affordable way to realize their full potential. It's at this point that Intel has decided to step in with the launch of its new Movidius module, which includes a vision processing accelerator focused on executing algorithms for object detection and facial recognition.
Google, Facebook, Microsoft, Apple, Amazon... one way or another, all these companies have various artificial intelligence projects in development. There are many reasons to follow this path, but one of the most relevant is cost reduction. Professor Moshe Vardi of Dice University has said that machines could take 50 percent of jobs in the next 30 years. We have already explored some of the possible effects in the past, and there are compelling examples like this factory in China that eliminated 70 percent of its workforce with robots. Now, artificial intelligence doesn't have to be exclusive to large companies. In fact, public intervention will be critical for its evolution, but first we need accessible tools. That's when Intel appears with its Movidius module.
History tells us that the company Movidius announced its Fathom deep learning USB stick in April 2016. Five months later, Intel acquired Movidius in a deal with confidential terms (the final sum never surfaced) and today Movidius arrives as a complete product from Santa Clara. Essentially, inside we find the Myriad 2 chip, a vision processing unit ideal for facial recognition and object detection tasks. Myriad 2 delivers more than 100 gigaFLOPs of performance with just one watt of power consumption, and can work natively with neural networks built on the Caffe framework (for now, no details on TensorFlow compatibility).
Initially, the module was priced at $99, but with its arrival on the market under Intel's wing, it will cost $79. Press images suggest it will be possible to combine multiple Movidius modules to further increase performance, however, the final decision will be up to each developer and researcher out there (graphics cards are powerful options, though they are more expensive and consume a lot of energy). Another possibility is that Movidius and the Myriad 2 could be integrated into multiple products. Smartphones, computers? Time will tell.
