If a patient is in critical condition, what and when should be measured to forecast detrimental events, especially under the budget constraints? This paper applies deep reinforcement learning (RL) to jointly minimizes the measurement cost and maximizes predictive gain, by scheduling dynamic measurements. The result was tested in a real-world ICU mortality prediction task, and reduced the total number of measurements by 31% or improve predictive gain by a factor of 3. This paper was presented in ICML 2019, and the authors are from University of Toronto and Vector Institute (An AI research institute led by Geoff Hinton).
Paper to read:
Chun-Hao Chang, M. Mai and A. Goldenberg, "Dynamic Measurement Scheduling for Event Forecasting Using Deep RL", ICML 2019
Data and code:
This is part of weekly reading series. We come together to discuss the cutting-edge AI topics and papers. One paper is selected each week as the major discussion topic.
You can attend in person or remotely. If you attend remotely, register here https://zoom.us/webinar/register/WN_lqDHHQQHTYK_QGPU3OJt5g
After registration, you will receive a link to join the presentation.
6:30-7pm Meet and greet
7-8pm Group discussion
8-8:30 Additional social