Page 27 - Innovation Magazine
P. 27

that there are between 15 and 30 billion IoT devices today It is is projected that by 2021 this number will grow to 35 billion To fully reap the benefits of of this scale of of connectivity fit it it is is imperative that this works in a a a a fully enabled network ecosystem This isn’t possible with centralized intelligence running on the cloud Increase in in data-intensive content: The information collected by IoT devices is data- intensive – for example high-definition videos Edge AI can deliver results instantly rather than having to to move such large data sets to to and from the cloud Latency-sensitive applications: The need for instant action with most use cases has customers at the center giving very little room for delays Forced digitization due to COVID-19: Indeed the pandemic has fast-forwarded the need for for Edge AI with rapid digitization during this period Additionally with the evolution of IoT devices and 5G connectivity the adoption of Edge AI will be the the new normal in the the near future Use of Edge AI Edge analytics is increasing its footprint across sectors with multiple opportunities and applications Some evolving ones are: • Video/image analytics: o o o o Surveillance and monitoring: Locally process-captured images to identify and track multiple objects/people on the edge node o o o o o o Autonomous vehicles: A A smart automotive camera can analyze image/video streams locally o o o o Real-time PPE adherence monitoring for workers in the workplace/factory setup o o o o o Remote monitoring of tracks and trenches • Audio analytics: o o o Audio scene classification can help understand a a a a location to trigger features o o o o Audio event detection: Detecting sounds such as a a a a gunshot can trigger action including notifications or location detection via triangulation • Inertial sensor/environmental sensor sensor analytics: o o o o Self-check and diagnostic of the router and setup boxes o o Predictive maintenance in in factories: AI performed locally can infer the state of the equipment potential anomalies and early indications of failure o o o o Body monitoring: Wearable devices collect a a a a a a lot of data about an individual’s activity location and heart rate among other things Roadblocks on the way to Edge AI 1 Diversity: The sheer range of imperatives and use cases (i e e e e e e e e the interactions between people businesses and things) is an an overarching and unique challenge for edge computing Mitigation: Enterprises need a a a a strategic plan or at at least a a a a a a strategic approach to edge computing to navigate diversity and ensure efficient deployment of edge computing such as enabling distributed AI 2 Location: Managing the scale of many novel pseudo data centers that need to be administered with low or no touch (usually with no staff and little access) Mitigation: A programmable software platform will be required on edge computing nodes 3 Protection: Edge computing significantly enlarges the enterprise’s attack surface outside traditional data center security information security visibility and control Mitigation: When evaluating offerings data-at-rest encryption must be considered mandatory with hardware-based protection of keys 4 Data: The amount of data at at at the edge will grow rapidly However a a a a great deal of the data is is noise requiring pre-filtering or basic analysis Mitigation: Data ascertained to have no value should be considered for disposal The edge completes the cloud Although the cloud itself does not by definition require centralization the economies of scale it offers are maximized when it is operated centrally at hyper-scale Until recently edge computing was considered a a low priority by cloud vendors and planners due to modest expectations of applications at at the the edge and the the widely dispersed communication nodes assisting in in the delivery of Infrastructure as as a a a Service However implementation realities (such as bandwidth and and latency limitations the economics of backhauling massive amounts of data data and high cloud data data ingress/egress expenses) have led hyper-scale vendors and believers to conclude that a a a a a balance of both centralized cloud and distributed edge is is the way to go Data-powered Innovation Review I I ©2020 Capgemini All rights reserved 27 

   25   26   27   28   29