Sensor Analytics at Micro Scale on the xPU
Written by Dr. Kirk Borne
We often think of analytics on large scales, particularly in the context of large data sets (“Big Data”). However, there is a growing analytics sector that is focused on the smallest scale. That is the scale of digital sensors — driving us into the new era of sensor analytics.
Small scale (i.e., micro scale) is nothing new in the digital realm. After all, the digital world came into existence as a direct consequence of microelectronics and microcircuits. We used to say in the early years of leveraging big data for astronomy, which is my background, that the same transistor-based logic circuitry that comprises our data storage devices (which are storing massive streams of data) is essentially the same transistor-based logic circuitry inside our sensors (which are collecting that data). The latter includes, particularly, the sensors inside digital cameras, consisting of megapixels and even gigapixels.
Consequently, there should be no surprise that the two digital data functions — sensing and storing — are intimately connected and that we are therefore drowning in oceans of data.
But, in our rush to crown data “big”, we sometimes may have forgotten that micro-scale component to the story. But not any longer. There is growing movement in the microchip world in new and interesting directions.
From CPU to GPU, FPGA and More
I am not only talking about evolutions of the CPU that we have seen for years: the GPU and the FPGA. We are now witnessing the design, development, and deployment of more interesting ASICs, one of which is the TPU which is specifically designed for AI applications. As its name suggests, the TPU can perform tensor calculations on the chip, in the same way that earlier generation integrated circuits were designed to perform scalar operations (in the CPU) and to perform vector and/or parallel streaming operations (in the GPU).
|ASIC||application-specific integrated circuit|
|CPU||central processing unit|
|FPGA||field programmable gate array|
|GPU||graphics processing unit|
|TPU||tensor processing unit|
Deep Learning: The Hot Trend in Machine Learning Algorithms
One of the hottest trends in machine learning algorithms for AI is deep learning. There are various implementations of these deep neural networks, one of which is TensorFlow™, which performs tensor calculations — essentially multi-dimensional matrix calculations — within the network to maximize classification performance of the algorithm. The TPU is designed to do tensor calculations in the firmware on the chip at the micro-logic level, thus greatly speeding up the calculations.
One of the most prevalent deep learning applications is computer vision, which is a field of computer science in which algorithms gain high-level understanding from digital images or videos. Computer vision has many applications, including:
- medical image diagnostics
- object detection and classification (within video and images)
- manufacturing and construction (from aerial surveillance or in situ imaging of an industrial site)
- weather and climate science (from satellite imagery)
- autonomous vehicles (from multiple streaming video cameras in the vehicle, sensing the environment and the physical space in which the vehicle is moving)
Autonomous vehicles especially require fast, nearly instantaneous, analytics — detection and understanding of the objects in the field of view of multiple cameras in a highly dynamic environment. Having a microprocessor chip in the camera’s sensor that can perform the computer vision calculations (object detection, recognition, and classification; and anomaly detection) in near real-time (at the point of data collection; i.e., at the edge) is an absolutely critical, potentially life-saving requirement.
Processing Behaviors of Interest
Speeding up the calculations is precisely the goal of these new chips. Other emerging examples are shown in the table at right. An interesting application that I heard discussed in the context of cybersecurity is the BPU, designed to detect behaviors of interest. Whereas the TPU might be detecting persons of interest or objects of interest in an image or in video frames, the BPU is looking at patterns in the time series (sequence data) that are indicative of interesting and/or anomalous behavioral patterns in sequences.
|BPU||behavior processing unit|
|EPU||emotion processing unit|
|IPU||intelligence processing unit|
|NPU||neural processing unit|
|SPU||stream processing unit|
|VPU||vision processing unit|
The BPU would definitely represent an amplifier to cybersecurity operations, in which the massive volumes of data streaming through our networks and routers are so huge that we never actually capture and store all of that data. So we need to detect the anomalous pattern in real-time before a damaging cyber incident occurs!
These capabilities are like a microchip analogy to a children’s detective board game from years ago – searching for forensic patterns in real-time streaming data to discover the “who, what, and where” of something significant that is happening – except that these data analytic patterns are much more interesting than the logic puzzles in a popular childhood board game.
xPU Implementations Moving Intelligence to the Edge
For simplicity, we can refer to the class of processors as the xPU, where “x” is now representing a very diverse and very interesting set of capabilities.
The development of new xPU implementations is driven by the demand for faster analytics on streaming data, powering and accelerating AI at the point of data collection — i.e., at the micro-scale, in the sensor — moving intelligence to the edge!
In addition to cybersecurity threat hunting and autonomous vehicle applications, other edge intelligence applications include location-aware time-specific hyper-personalization in consumer marketing, unit-level supply chain item identification and tracking, wearable devices on employees in high-risk jobs, real-time video and audio sentiment analysis of persons (including emotion detection and triage decision support in emergency response scenarios), and 4-D printing.
As more and more sensors proliferate ubiquitously for many beneficial purposes in our workplaces, buildings, industrial sites, highways, cities, farms, skies, personal health wearable devices, and homes, the need for edge analytics (computing at the edge) is similarly growing rapidly. Similar to the childhood board game analogy mentioned earlier, “at the edge” really means “sensor analytics at micro scale on the xPU”.
How IoT and Edge Computing Solutions are Extracting Real-Time Insights
TensorFlow is a trademark of Google Inc.