Cross-platform deployment and update of neural networks

Cross-platform deployment and update of neural networks

.
.

Background

Nowadays AI is a hot key word in many application fields as well as in the academic world. Designing, training and evaluating deep neural networks are must-to-have skills for people in this domain. However, there are fewer concerns about how to deploy the DNNs on real product and maintain their functionality and efficiency via seamless update.


Usually developing a DNN requires high-end hardware to make it possible. But such cost is not realistic for the end users in most cases. There might be lack of GPU and only low-cost devices are available. In certain use cases the DNNs should be deployed on mobile or embedded devices to meet the customer requirements. Nevertheless such service should be sometimes accessible via Web or Cloud service for better user experience. Besides, regular updates of DNNs are also a crucial part.


Tasks

Under this topic you will look at the state-of-the-art technologies for cross-platform deployment of AI solutions. Given a specific application field you will develop a whole architecture to manage, deploy and update DNNs in a highly automatic way. The DNNs should be converted, compressed and adapted to different platforms and a seamless update across platforms (e.g. OTA) should be achieved.


Required skills

  • High motivation to learn new technologies
  • Skills in programming (Python, C++, C, Java ...)
  • Knowledge about network and hardware
  • Experience in machine learning field is a plus