Question $3$: Create a video presentation to share your thoughts on one of the following topics $($video should not be more than $5$ minutes long$)$:

Topic $1$: How can Language Vision Models be made more robust to domain shifts and data distribution changes across diverse real$-$world environments, while maintaining efficient inference on resource$-$constrained devices?

Topics $2$: How can large vision models be enhanced to generate high$-$quality, diverse, and contextually coherent images while reducing the impact of spurious correlations in the training data?