Most of today’s IoT predictive analytics rely on a cloud-centric architecture in which embedded edge/gateway devices send continuously the collected data to the cloud, this in order to carry out the prediction process and eventually update the predictive model. However, this architecture causes different problems like for instance (i) large amount of irrelevant data that is sent and stored into the cloud (ii) high energy consumption of edge/gateway devices caused by continuous (wireless) transmission which decreases their battery lifetime (iii) loss of data privacy and security by releasing localized sensible data to the cloud.
Recent advances in embedded processors enable new opportunities like executing analytics at the edge and/or gateway devices, therefore enabling truly real-time and distributed predictive analytics. One of the challenges of edge/gateway-centric architecture is how the embedded predictive model is regulary updated and personalized with respect to the monitored context, by taking into account real-time collected data. This challenge constitutes the basis of the proposed internship.
Work to be performed by the trainee
The first part of this internship will focus on a theoretical analysis of some pre-selected embeddable machine learning (ML) techniques like Support Vector Magnitude (SVM) or Neural Networks (NN). Next, the student will implement at least one selected ML algorithm to be executed on a lightweight embedded platform such as Raspberry PI or Arduino. The predictive model will be tested and evaluated in the context of monitoring a human activity at home.
The ideal candidate for this internship should have good programming skills in C/C++ and linux and have basic knowledge of machine learning techniques.
The internship is scheduled to start after February 1st, 2016 for a duration of at least 3 months. Interested candidates can submit their CV and an internship cover letter to Philippe Drugmand
keywords: real-time predictive analytics, embedded machine learning for IoT