Last semester, I took the class on “Embedded Intelligence” with Prof. Lukowicz. It was the first offering of the course, since the group has moved recently from Passau to DFKI,Kaiserslautern.
It was relatively a newer area of technology for me to explore and it was fun knowing a lot of concepts.
Embedded Intelligence (EI) falls between the realm of Classical AI, where the goal is to develop systems that can “understand” and that of Embedded controllers, involving simple feedback systems. The techniques developed in EI involve using the available infrastructure of urban living, to make sense about user state and activities. (For eg., using sensors in smartphones to know about user movement and daily activities.) A central concept is that of a “Digital Shadow” – data trail left by any real world activity in the digital domain.
Where Embedded Intelligence would thus cater to individual user perspective, ‘Socially aware computing‘ would extract high level information from infrastructure around multiple users, and in turn, dynamically adjust system configuration. (For eg., crowd monitoring in mass gatherings using location data from cell phones. Suggesting alternative routes to users in a crowded area.) Smart infrastructure would include all personal digital devices and those set up in the surroundings.
And here are some wisdom crumbs to take away :
- If you torture your data enough, it will tell you anything. (Prof. would say this in almost every class. =) ) Make sure data isn’t too distorted.
- Its hard to be objective even in Sciences like Physics.
Story : Millikan’s oil drop experiment. The value of the coefficient calculated by Millikan in the original experiment was incorrect. But it took quite sometime until the correct value was established in the scientific community. Values calculated by each subsequent experiment differed by a little to the previous value until the true value was reached. Researchers thought it would be safe to differ just by a little. (even when the indications of the experiment were quite different.!)
(Though personally I might not quite agree. I remember how I would do all experiments back in school,even after everyone else would have left the lab; who cared about the results when doing experiments was the fun. =) )
- Co-existence does not imply correlation.
Story : There was a study that said :The rate of crime is proportional to the number of churches within that area!
What occurred was there would be lots of churches in downtown area of a city, where there would also be most of the residence and hence also more the no. of crimes. But clearly,there is no relation!
- Sensor Fusion : Choose sensing modalities whose sources of errors are statistically independent. (Very important lesson when combining decisions from different sensing modalities for any decision making.)
- Understand the data before applying any Machine Learning techniques. You must know how your data is in different conditions and what changes it and how, before any processing is done.
Course was super fun and so were the exercises. I would recommend it to anyone interested in future ICT and in general in impact of technology on society.