Because efficiency is key to this challenge, any solutions that use brute-force grid search and static machine learning that do not utilize the simulation engine for inference will not be eligible for final submissions.
I have two questions regarding these rules:
-
Why does static machine learning conflict with efficiency? Generally, iterative algorithms incorporating simulation engine might be more time-consuming.
-
What specific types of static machine learning algorithms are being referred to? Does this include end-to-end deep learning models?
Hi @shaodong hopefully this helps:
We are looking for efficient solutions that integrate the simulation engine. If you have a machine learning model (including deep learning) that integrates the simulation engine at inference time, that is exactly what we are after. What we are looking to avoid are brute force and massive dataset generation solutions. Brute force’s inefficiency is pretty straightforward. With a massive dataset it might be efficient at inference time, but building and training on a dataset tends to be very time consuming.
Happy searching!
ThinkOnward Team