The success of machine learning algorithms depends heavily on the representation of the data. It is widely believed that good representations are distributed, invariant and disentangled. This challenge focuses on disentangled representations where explanatory factors of the data tend to change independently of each other.
Given the growing importance of the field and the potential societal impact in the medical domain or fair decision making, it is high time to bring disentanglement to the real world:
Stage 1: Sim-to-real transfer learning - design representation learning algorithms on simulated data and transfer them to the real world.
Stage 2: Advancing disentangled representation learning to complicated physical objects.
For more details on the challenge and how to participate, please consult the Disentanglement Challenge website: