Seminar: Resource-Aware Context Recognition using Deep Learning on Mobile Systems
The aim of the seminar is to design knowledge extraction algorithms based on deep learning techniques for activity recognition in free-living using wearable sensing systems. Challenges will be related to imbalance of target classes, scarcity of computational resources and design of parallel computing coding structure. Evaluation will be performed on free-living dataset and benchmarked in terms of energy/memory savings, time constraints and quality of information retrieved. Prototyping will be done in Python using Keras and PyTorch libraries.
- Gain overview on the state-of-the-art of wearable technology for activity recognition.
- Gain overview on the state-of-the-art of software-based power management for wearable devices.
- Learn how to exploit deep learning for signal processing and data abstraction.
- Learn concepts of sparse sampling.
|ECTS||2.5, 5, 7.5, default: 5|
|Project type||Seminar, Extension to BSc.MSc.-Thesis can be discussed|
|Presence time||4 SWS|
|Work distribution||25% Theory, 75% Programming|
|Useful knowledge||Python programming, machine learning basics|
|Period||Summer semester 2019|
|First meeting||Seminar introduction/Vorbesprechung on
24. Apr 2019, 17:00-18:30 at Henkestr. 91, Haus 7, 1. OG, R 373
|Med. Eng. Seminar Title||Advanced Context Recognition (ACR)|