This video is #11 in the Adaptive Experimentation series presented at the 18th IEEE Conference on eScience in Salt Lake City, UT (October 10-14, 2022). In this video, Sterling Baird @sterling-baird describes mixed online offline multi-fidelilty optimization or scenarios where humans have to go into the lab and collect data guided by simulations. For example, an optimization problem might be solved offline using previously collected data, and then the solution is refined online using real-time data. This approach can be useful when the data needed to solve the optimization problem are not available in real-time, but can be collected and used to find a rough solution offline. The online component can then be used to fine-tune the solution based on current data. Our next video covers closed-loop experimental optimization.
Github link to jupyter notebook [ Ссылка ]
previous video in series: [ Ссылка ]
next video in series: [ Ссылка ]
0:00 defining mixed online offline optimization
1:40 video overview
2:45 scheduler API
4:48 scheduler class
6:05 setting up mock external execution system
9:55 setting up experiment
14:10 define callback function and running optimization
16:06 configurations for scheduler
17:12 advanced functionality (report results to external system etc)
