Recently, 3D point cloud deep learning has been making remarkable strides in various domains, e.g.,

The pre-trained Large Language Models (LLMs) can be adapted for many downstream tasks and tailored to align with human preferences through fine-tuning.

On one hand, it operates as a general data augmentation strategy.

Any explanation that faithfully explains this type of model needs to be in agreement with this invariance property.