The goal is to train a tiny deep neural network that performs Image Segmentation and Optical Character Recognition (OCR). It translates any sections of a screen into continuous text. Existing networks are based on Convolutional Neural Network (CNN). The thesis comprises two milestones (A) Creating a baseline system. (B) Reduce the memory requirements of the CNN. This can be done by reducing the number of parameters. The recognition accuracy is maintained by distilling knowledge from a previously trained teacher network. This internship or thesis project has a focus on machine learning. You will need to work with Pytorch on powerful multi-GPU server.
Please note that students need to be enrolled during the whole period of employment in order to be eligible for the student internship /thesis project.
For this student Position we are preferably looking for a student to join our team from Spring 2021 for a duration of 4-6 months
- Studying towards a Masters degree in Artificial Intelligence or similar discipline, also possible for PhD students
- Good working knowledge of algorithms and compute structures in deep learning and artificial neural networks especially inference
- Good working knowledge of speech audio and language processing techniques used in recognition classification identification and/or translation systems
- Good working knowledge of computer architecture and algorithm optimization
- Software programming skills CC Perl Python are used in our group
Inside this Business Group
IP Engineering Group's (IPG) vision Build IPs that power Intel's leadership products and power our customer's silicon. We want to attract & retain talent who get joy in building high quality IP and share our core belief that IP is fundamental to transforming Intel's silicon design process. IPG's guiding principles will be ensuring Quality (Zero Bugs), Customer Obsession (Delight our Customers) and structured Problem Solving. We are a fearless organization transforming IP development.
DE Intern JR0158343 Munich IP Engineering Group (IPG)