Learning object-centric representations of complex scenes is a promising step towards enabling efficient abstract reasoning from low-levelperceptual features. Yet, most deep learning approaches learn distributed representations that do not capture the compositional properties of natural scenes.

Existing approaches based on off-policyevaluation (OPE) oftenrequireadditional function approximation and hence hyperparameters, creating a chicken-and-egg situation.

The recent large-scale text-to-image generative models have attained unprecedented performance, while people establishedadaptor modules like LoRA and DreamBooth to extend this performance to even more unseen concept tokens. However, we empirically find that this workflow often fails to accurately depict the out-of-distributionconcepts. This failure is highly related to the low quality of training data.

Parameters5 10 3 5 10 3 5 10 3 (Critic Update:Environment Step) Ratio 1 1 1 (Policy Update:Environment Step) Ratio 2 2 1 Has Target Policy?

While this could be done using thesynthesis formulation, we demonstrate that this leads to slower performances. The main difficulty inapplying suchmethods intheanalysisformulation liesinproposing a way to compute the derivatives through the proximal operator.