Learn Baby Learn
True, Artificial Intelligence (AI) and Machine Learning are no longer in infancy, but the excitement and burst of fresh ideas and applications happening now has all the earmarks of a new arrival in the world.
Technology constantly challenges how we traditionally operate. However, with great change comes tremendous opportunity. Meet the ascension of affordable Artificial Intelligence (AI), big data, and lots of processing power for Technology’s Next Big Thing.
As Diane Bryant, Executive Vice President of the Data Center Group at Intel® pointed out at IDF 2016, “This is a period of tremendous opportunity. The future is cloud, connectivity, and artificial intelligence.” At the forefront of AI is machine learning; a way to analyze data that automates the creation of new analytical models. Machine learning does not require programming like traditional computers, but rather iteratively “learns” by examining from dozens to thousands of sample sets that demonstrate decision-making much like human reasoning. Machine learning can quickly find insight from massive, messy data sets. Technology has us at a crossroad where large amounts of data, low cost of data storage, and low-cost, high-performance computing intersect, creating accessible advanced analytics.
AI is growing more pervasive in our daily lives as it is used for fraud detection, security video monitoring for behavior anomalies, pattern matching and recognition,online advertising, network intrusion detection, predicting equipment failure, and text-to-speechconversion (see the article by senior editor Caroline Hayes in this issue, “Intel® Natural Language Processing Won’t Hit the Wall Anytime Soon,” in which she tells us how “Oakley’s Radar Pace” eye wear relies on Intel’s Real Speech technology for contextual, natural-sounding two-way coaching.”)
AI is often applied to mind-numbingly boring work that finds humans searching for more fulfilling outlets. With AI, casinos can monitor dozens of security feeds for anomalous behavior from gamblers that alert human security personnel for detailed real-time examination. AI security monitoring can alert security personnel to cars or people going the wrong way in access/egress channels, 24/7, from hundreds of available monitors without ever getting bored or tired.
Most machine-learning tasks are intensively computational. Data analytics is the fastest growing workload for servers, growing faster than any other server workload. “By 2020, more servers will be running data analytics than any other type of workload. Intel processors power 97% of all the servers deployed for machine and deep learning today,” according to Bryant.
To meet the demand, Intel’s premier architecture for AI is the Intel® Xeon Phi™ processor, which delivers scalable performance for deep learning, as models move from sample learning sets to massive amounts of real data. Machine learning requires a parallel and distributed high-performance architecture. GPUs are good at parallel processing, but are not as scalable as CPUs, and here is where Intel excels. Offloading to a GPU can be suboptimal, especially when hundreds of individual models are the way you do analytics for customers, according to Indico CEO and Founder, Slater Victoroff. Indico, a text- and image-analytics company, uses “transfer learning” to accelerate model-making, fine-tuning machine learning into hundreds of models tailored for each customer. GPUs, although in wide use, require extra “hops” when dealing with a GPU and a CPU, or figuring out the communication pathways for each model. A GPU is also low in memory. It makes more sense to use a CPU with a healthy memory footprint when you want to scale like Indico has done.
AI can take on menial “mental” work, much like steam powered machines replaced manual labor in the industrial revolution. Call it Industry 4.0, call it evolutionary or revolutionary; whatever the name, it’s going to change our lives in numerous ways.