The challenge in object-based visual reasoning lies in generating concept representations that are both descriptive and distinct.

We also evaluate our algorithm empirically. By leveraging a state-of-the-art nearest-neighbor search library, we obtain a fast and accurate Centroid-Linkage HAC algorithm.

In the absence of class priors, recent deep clustering methods resort to data augmentation and pseudo-labeling strategies to generate supervision signals.

We explore those different aspects in relation to mutual knowledge, and analyze zero-shot predictions.

Meanwhile, diffusion models emerged as a powerful framework for multimodal learning.