What is Machine Learning on the Edge?

Machine Learning on the Edge, or Edge ML, is an AI technique that enables IoT smart devices and applications to process data locally at the source through machine learning and deep learning algorithms. It is referred to as Machine Learning on the Edge because data processing occurs locally, either on the device or at a server level, via machine learning algorithms.

Anyone responsible for developing an intelligent connected device should consider an edge computing solution for the following reasons:

  • Latency: Internet and the cloud are too slow for many intelligent connected device applications
  • Remote Locations: There may be poor or unreliable internet connection
  • Form-factor: Edge devices can be physically much smaller and consume less power than Cloud TPUs
  • Privacy: Images and potential sensitive sensor data does not need to be sent off the device, only the meta data and insights generated need to be sent.
  • Cost: Reducing data transfer to the cloud may reduce overall cost

 

The use of Machine Learning on the Edge for IoT smart devices reduces reliance on cloud networks that sometimes become congested with data and experience high latency. Edge ML can analyze large amounts of data to determine if it should be processed in a local data centre or in the Cloud. Because only the final result is transmitted to the data centre, Machine Learning on the Edge can lower latency and conserve bandwidth.

Machine Learning on the Edge also offers solutions regarding storing personal user data thanks to the option of a local server rather than in the cloud, which can be vulnerable to security breaches.

 

What are the advantages of Machine Learning on the Edge over the Cloud?

Machine Learning on the Edge enables the processing of data in real-time, which is currently not possible with traditional, cloud-powered IoT devices, yet critical for technologies like autonomous vehicles and medical devices that must process data quickly and efficiently.

 

The increase in connected devices means an increase in potential security breaches

 

Intelligent connected Edge devices can make continuous improvements and updates after they are deployed in the field, as a central service aggregates the new data and then uses Machine Learning algorithms to implement the updates to each device. Many industries have adopted Machine Learning on the Edge systems into their applications in part thanks to its higher latency and better security than the cloud. Some common use cases of Machine Learning on the Edge that span across multiple verticals include:

  • Autonomous vehicles using Computer Vision
  • Manufacturing and Industrial predictive maintenance, such as:
    • Quality control by inspecting for defects
    • Easy additions of quality assurance equipment to existing manufacturing lines
    • Spectral analysis for faults, food inspection, and pick-and-place
    • Predictive Machine Learning sensors and algorithms for monitoring of system health and machinery that notify technicians for maintenance
    • Systems shut down in case of dangerous malfunction
  • Healthcare
  • Virtualized radio networks and 5G (vRAN)
  • Cloud gaming
  • Smart grid
  • Content delivery, such as music and video
  • Traffic management
  • Smart home automation, such as:
    • Maintenance (light bulbs, fuses, etc)
    • Inventory planning (smart fridge, etc)
    • Home assistants (fashion advice, elderly care)
  • Smart retail
    • Machine Learning algorithms identify trends in customer behaviour
    • Inventory management
  • Recreation
    • Virtual coaches for various sports activities (golf, tennis, swimming, yoga, etc)
    • Safety monitoring (climbing, swimming, skiing, etc)

 

What engineering expertise does MistyWest offer in Machine Learning on the Edge?

MistyWest’s skills and expertise in Machine Learning on the Edge includes:

  • Hardware accelerators, such as FPGA
  • Machine Learning enabled microprocessors (Machine Learning on ST Microelectronics IMU platform, etc)
  • Hardware accelerators on embedded Linux: NVIDIA GPUs on the NVidia Jetson platform, Google Coral, Intel Movidius
  • NVIDIA accelerators: DeepStream, Issac
  • Containerized Machine Learning application deployment to embedded Linux
  • Tensorflow
  • Tensorflow Lite for Microcontrollers, TensorRT for GPU inference optimization, OPenVino for intel architectures

 

The adoption of Machine Learning on the Edge into IoT engineering covers a wide range of businesses, from hardware startups developing their first intelligent connected devices, to Fortune 500 fashion companies that are pivoting to wearables and higher technologies, to large companies conducting exploratory research and development.

MistyWest’s clients have included some of the above, and the types of Machine Learning on the Edge projects are always varied. Some of MistyWest’s previous work experience in Machine Learning on the Edge includes:

  • Docker containers optimized for NVIDIA Jetson.
  • Machine Learning based Docker containers for Fortune 500 companies
  • A wearable with a small microprocessor with Machine Learning based gesture detection for an Alphabet company’s experimental R&D division
  • Edge-based speech to text with a live display of the speech

 

We hope this FAQ gives you a greater understanding of what MistyWest’s Machine Learning on the Edge capabilities are and how you can leverage the Edge in your intelligent connected device development.

If you’re interested in learning more about MistyWest’s Edge ML services, request one of our detailed capabilities statements or send an email to [email protected] to be put in touch with one of our business developers.