Description de l'offre
Master Thesis within Artificial Intelligence/Neuronal Network performance optimization
Airbus Defence and Space Manching
Airbus is a global leader in aeronautics, space and related services. In 2017, it generated revenues of € 67 billion and employed a workforce of around 130,000. Airbus offers the most comprehensive range of passenger airliners from 100 to more than 600 seats. Airbus is also a European leader providing tanker, combat, transport and mission aircraft, as well as Europe’s number one space enterprise and the world’s second largest space business. In helicopters, Airbus provides the most efficient civil and military rotorcraft solutions worldwide.
Our people work with passion and determination to make the world a more connected, safer and smarter place. Taking pride in our work, we draw on each other's expertise and experience to achieve excellence. Our diversity and teamwork culture propel us to accomplish the extraordinary - on the ground, in the sky and in space.
Description of the job
Are you looking for a master year project? Would you like to discover the work of Combat Aircrafts? Then apply now! We look forward to you joining us at Airbus Defence and Space. +
Start: Spring 2019
In the area of the Combat Aircrafts Airbus developed a real-time capable neural network solution. Neural networks for real time applications require considerable amount of hardware resources. Your task is to find possibilities to optimize the existing neural net solution in a way that the existing solution requires less hardware resources while keeping current network quality and performance.
Tasks & accountabilities
The thesis would be divided into three parts:
In a research part you would concentrate on already existing neural network optimization techniques to find out if and how these can be adapted to our context. NVIDIA and Intel for instance offer frameworks for this purpose (TensorRT, Intel Deep Learning Inference Accelerator). In the field of computer vision there are specific neural network architectures which are optimized for Real-time Capabilities (e.g. YOLO or Single Shot Multibox Detection).
Following the research and selection phase the optimization opportunities shall be further amended with your own ideas and an implementation shall be developed on the target platform.
Finally a before/after case benchmark shall be carried out based upon selected criteria like network size, memory consumption or CPU-load to measure the objectives of the master thesis with the help of the applied solutions.
This job requires an awareness of any potential compliance risks and a commitment to act with integrity, as the foundation for the Company’s success, reputation and sustainable growth.
· Enrolled student (m/f) within Computer Science, Electrical Engineering, Data Science or similar field of study
· Programming experience in C++ / Ada / Python
· Experience in one or more of the following deep learning frameworks as well as experience in the area of machine learning in general:
· Background in the context of embedded hardware development
· Knowlege in the following areas is an advantage:
· Nvidia Deep Learning SDK
· Personal skills:
· Independent analytical thinking
· Good communication skills in German or English