Neuroevolution

Problem Difficulty Classification

Difficulty Mode

Training Set

Testing Set

easy

Deep networks (depth > 2)

Shallow networks (depth ≤ 2)

difficult

Shallow networks (depth ≤ 2)

Deep networks (depth > 2)

Note: Total 66 networks available. When difficulty is ‘all’, both sets contain all networks.

This problem set is based on the neuroevolution interfaces in EvoX. The goal is to optimize the parameters of neural network-based RL agents for a series of Robotic Control tasks. We pre-define 11 control tasks (e.g., swimmer, ant, walker2D etc.), and 6 MLP structures with 0~5 hidden layers. The combinations of task & network structure result in 66 problem instances, which feature extremely high-dimensional problems (>=1000D).