Universal Feature Representations

Sayak questions the idea of universal visual representations for different domains and tasks, suggesting more focus on learning adaptable features. Tim emphasizes fine-tuning models in-domain for better results, highlighting the importance of domain-specific training over universal representations.