The global artificial intelligence output is expected to reach 200 billion US dollars by 2022. If the current trend continues, health care will occupy a considerable share of this market. This is not surprising in terms of the potential of artificial intelligence technology. Artificial intelligence can effectively reduce management costs, reduce patient waiting time, and conduct self-diagnosis of diseases. Today, Intel and Philips show two other applications of artificial intelligence: bone modeling and lung segmentation.
The Philips Medical Supply and Sensors Division published the results of one of its most recent machine learning tests. This test was performed with Intel's Xeon Scalable processor and was complemented by the OpenVINO Computer Vision Toolkit. The researchers focused on two use cases: the first is to use bone X-rays to simulate changes in bone structure over time, and the second is to use lung CT scans to segment the lungs (ie, to determine the boundaries between the lungs and surrounding tissues).
They successfully increased the scanning speed of the bone age model by 188 times, from the baseline of 1.42 images per second to 267.1 images per second. The speed of the lung segmentation model has increased by a factor of 38, from the previous 1.9 images per second to the current 71.7 images per second.
Vijayananda J., Chief Architect, Philips HealthSuite Insights “Intel's Xeon Scalable processor is clearly tailored for this type of artificial intelligence workload. Our customers can use this processor to maximize their potential hardware while achieving high speeds at very high speeds. Quality output resolution."
Intel says its processor is not just a powerful graphics card for training and running machine learning models. It has a key advantage in computer vision: it can handle larger, more memory-intensive algorithms.
In May of this year, Intel claimed in a blog post that its Xeon platform outperformed NVIDIA's Volta 100 in terms of machine self-inference tasks (such as machine learning translation). And Intel recently published a case study of Novartis, which showed that Xeon's image analysis model for early drug discovery was more than 20 times faster than before.
One thing is clear: Intel is preparing for the growth of its artificial intelligence chip business. In August, the company announced that it has sold more than 22 million Xeon processors in the past 20 years, generating $130 billion in revenue. This is a far cry from the artificial intelligence market's expectation of $200 billion in 2022, but the company plans to narrow the gap with expectations by gaining $20 billion in market share over the next four years.
Of course Intel is capable of doing this. The chip giant previously acquired Altera, adding a field-codeable gate array (FPGA, an integrated, reconfigurable circuit) to its lineup. In addition, several other recent acquisitions—that is, Movidius and Nervana—have also enhanced Intel's real-time processing business. It is worth mentioning that Nervana's neural network processor is expected to start production in late 2019, and its artificial intelligence training performance is reported to be 10 times that of competitive video cards.
In addition, Intel said its upcoming 14nm Cascade Lake architecture processor will be 11 times better than the previous Silver Lake platform in image recognition, and the processor will also support a new type of artificial intelligence. Focus on the instruction set, called DL Boost.
At the data-centric innovation summit held this month, Intel executive vice president Navin Sheno said: "For 50 years, this is the biggest opportunity for the company. We now occupy 20% of the market, our The main strategy is to promote the arrival of a new era of data center technology."