Lecture Embedded Intelligence
Here you can find the information about my lecture Embedded Intellgience that is taught in the University of Kaiserslautern
Since the lockdown measures from Covid-19 pandemic, the lecturers from the University of Kaiserslautern have moved our contents online so that students can study remotely. This has given me the perfect opportunity to summarize and customize the content and make it available for everyone. All materials are for educational purposes. If you are the owner of some materials from the content and do not wish them to be there, please contact me.
Summary
This lecture introduces various methodologies of human activity recognition using embedded (wearable or ambient) sensors. This lecture is a bridge between embedded systems and machine learning. Usually half of the class are from the electronics department and the other half are from the computer science department. In this lecture we will learn from how IMU sensors are made, to how to use deep learning to detect physiological and psychological activities. This lecture focuses more on the idea and flexible ways of thinking while solving problems. I try to inspire students to jump out of the box and look at the problems from different perspectives.
Syllabus
Click on individual lecture for details.
1. Introduction to Embedded Intelligence
The vision of embedded intelligence, that computing should be indistinguishable from everyday objects. Introduction of activity recognition. The categories and hierarchy of activity contexts. The ethical topic in artificial intelligence.
2. Localization
Location awareness is the fundemental context in embedded intelligence, including topics from localization to locomotion tracking. We will introduce the basic localization principles, satellite locationing systems, indoor localization techniques, and location context models.
3. Wearable Motion Activity Recognition
Activity and gesture recognition using IMUs. Introducing machine learning workflow, time signal processing, frequency analysis with fast fourier transform and wavelet transform, and result presentation with confusion matrix.
4. Ambient Intelligence
Ambient activity recognition with multiple environment sensors. Introducing sensor fusion, classifier basics, overfitting and spotting.
5. Wearable Biomedical Sensing
Monitoring physiological signals with embedded sensing systems. Indepth look at using surface pressure mechanomyography to detect various muscle activity patterns, including introduction to dynamic time warping.
6. Activity Recognition and Deep Learning
Introduction to neural networks and deep learning in the context of activity recognition. Explains dense and CNN layers. Further discussion on sensor fusion and neural networks with the RBK dataset study.
7. Transfer Learning in Activity Recognition
Knowledge domain transfer utilizes part of models trained with other knowledge to detect different domains. Further introduction on RNN units. Examples with image networks in pressure mapping imagery recognition, and time series recognition with spectrogram.
8. Mental State Recognition
Affective computing and psycho-informatics. Examples of bipolar detection using smartphone methods, expression and cognitive load detection using Expressure, and affective touch for robot skin.
9. Optimization of Embedded Intelligence Systems
Optimize the computing and power efficiency for embedded sensing and activity recognition systems. Detailed analysis on feature selection and reduction using PCA, LDA and NCA methods.