According to The Economist Intelligence Unit IoT Business Index 2017 (Link), though 90% of business owners accept the evolving convergence of IoT in business, there are 2 major deterrents of wide-spread business implementation of IoT Technology:
1. High cost of required investment in IoT Infrastructure,
2. Security and privacy challenges
In order for a business to reap value from connected devices over internet, the crucial missing piece which needs to be harnessed is the right utilization of the humongous data from the ever increasing number of connected devices in terms of time sensitivity as well as bandwidth of the data being transferred. This is most likely to be possible when the following basic conditions, which reduces the cost and increases the return from an IoT implementation, are met:
> Limited human intervention
> Limited hardware
This is when the concept of Edge processing comes into picture. Edge processing (or Fog Computing, as Cisco originally coined the term) is the concept which in simple terms uses data from unlimited sources before it looses significance as it brings processing at the edge of the network or ‘on-the-device’. It is not a new idea as reducing the dependency over the cloud and internet has been understood since the importance of cloud infrastructure was understood in real scenarios and would be an important piece in commercializing the connected devices in the future. Example: India’s leading ride sharing operator Ola launched its Ola Offline App in the late 2016. With consistently increasing number of connected devices, one can only think of the ease edge processing can bring to an IoT solution both it terms of the RoI of an IoT investment as well as Security. This means devices to turn and act ‘intelligent’ on their own and hence the intersection of artificial intelligence.
USE CASES which see a huge amount of data generated by IoT / mobile devices can definitely use the high precision machine learning / deep learning (ML/DL) platform as it can tackle some of the major issues related to data like: Latency, Bandwidth, Security of the data, as well as the Cost incurred in transmitting it, apart from several others. Some of the use cases which can be thought of are:
1. Time sensitive decision making in connected devices — Decision making in autonomous vehicles where information processing needs to be done in micro/milli seconds and it can’t afford the transmission lag and latency issues
2. When enormous data should be processed before being transmitted like monitoring plants in a vast farmland, where only the information about withering or infected plants needs to be transmitted based on certain sensor parameters
3. Security is of concern in automated industrial plants or surveillance cameras where local processing and intelligence of huge amount of recorded data is of high priority
4. Smart Home applications which need reduced human intervention like (taking inspiration of Amazon Echo here) replenishing the empty food bottles and cans, or switching between devices might also use edge analytics use case
A CHALLENGE seen in the intersection IoT with AI can be the fact that high precision ML/DL platform usually require higher processing and costlier hardware. The use cases listed above and similar upcoming use cases might depend more on how the technology solves the the size and cost of processing in the edge devices.
RECENT EXAMPLES: NVIDIA’s Jetson TX2 launched in March 2017 claims to be the 1st supercomputer on a module, a processor to bring AI to the Edge. Another example being that of Qualcomm’s deep learning SDK for devices called Zeroth Machine Intelligence Engine SDK which enables running of independent OEM’s neural network models on its Snapdragon 820 devices such as smart phones, security cameras, automobiles and drones, without any dependency to the cloud.
This post is inspired from my quora answer which first appeared here