Machine Learning for IoT at the Edge

Description:

With billion of device getting connect and generating petabytes of unstructured data, the need to have near real-time insight has also become crucial for various industrial applications. In this talk we will particularly demonstrate why edge computing is crucial to IoT in deriving the essential ROI of IoT applications extending it to a case study. The outline of the talk will be as follows -

What is edge computing and its architecture and the beneficial features it delivers (2 mins)

• Speed

• Scalability

• Guaranteed reduced latency and guaranteed effective bandwidth

• Privacy and Security

• Last but not least, reduced cost

Why is edge analytics crucial to IoT: discussing specific use cases of various industry verticals (5 mins)

Fog computing, how is it different from edge computing and how is it enabling edge computing. Extending beyond, how is edge computing complimenting the existing cloud computing infrastructures (3 mins)

Various machine learning and deep learning architectures available for resource constraint analytics (5 mins)

Demo: (10 mins):

Consider a case study of healthcare industry, a patient suffering from brain injury is being monitored in an ICU 24*7 with the help of connected healthcare equipment's where the readings of his body vitals are collected every second. Each second is crucial to the patient's life. These vitals collected are sent to the cloud to analyse for occurrences of seizures or any other extreme events related to internal damage within the brain. This to and fro travel of data and information are dependent on the network bandwidth, which is time consuming as well as vulnerable to cyber attacks. Medical data are always sensitive data. At a situation where the network bandwidth or security is compromised, we would need an infrastructure which is capable of delivering immediate insight closer to the sensors collecting the data so that the required action can be performed immediately as well as privacy and security of the data is maintained.

Keeping this scenario in mind, a seizure prediction solution will be demonstrated on Raspberry Pi (resource constraint edge device) using machine learning and AWS IoT capabilities to derive secured, minimum latency insights offline without depending on network bandwidth and the time lag between to and fro travel of data and information between sensors and cloud.

• Overview of the AWS Greengrass for IoT edge analytics (1 min)
• Walkthrough of the solution architecture (3 mins)
• Solution setup (2 mins)
• Code walkthrough (4 mins)

Note: all code demonstration will be in python and its stack of libraries

Key Takeaways:

• The key takeaway from this talk will be an understanding of the edge infrastructure for IoT and how is it working hand in hand with the cloud computing infrastructure as well as why it is becoming crucial to actually put IoT data to work in deriving the correct insights.
• Further ahead, with the demonstration, audience's will get to understand how an enterprise level IoT analytics solutions can be architected and developed at the edge.

Prerequisites:

• Basic knowledge of Internet of Things
• Basic knowledge of Machine Learning

Content URLs:

Slides deck : https://drive.google.com/drive/folders/10ms7HVuCA2Sd3WJFK4IrWkpfam-G335_?usp=sharing

Speaker Info:

Sukanya is a Data Scientist working with Capgemini. She has extensive experience working with IoT building various kinds of solutions. She enjoys the most when she works on the intersection of IoT and Data Science. She also leads the PyData Mumbai and Pyladies Mumbai chapter. Besides work and community efforts she also loves to explore new tech and pursue research and has published a couple of white papers with IEEE and a couple more are in the pipeline.

Speaker Links:

1. Linkedin: https://www.linkedin.com/in/sukanyamandal/
2. PyData Mumbai: https://www.meetup.com/PyDataMumbai/