Questions for "Scaling Vision Transformers to 22 Billion Parameters" ----------------------------------------------------------------------------------------------- Please send your answers to: jesslen@cs.uni-freiburg.de by 13:30 on 05.07.2023 1) Briefly describe the measures taken to mitigate training instabilities. (~2-3 sentences) 2) In your opinion, what is the most noticeable improvement of ViT-22B from previous smaller ViT variations? Why? (~2-3 sentences) 3) How can we leverage large models such as ViT-22B to enhance the accuracy of existing models, and could you provide a brief explanation of how this process works? (~3 sentences)