Questions for "End-to-End Learning of Visual Representations from Uncurated Instructional Videos"
-----------------------------------------------------------------------------------------------

Please send your answers to: bravoma@cs.uni-freiburg.de by 10:15 on 30.07.2020

1. Explain the principal obstacle/problem the authors tackle while working with videos "in the wild", and why their MIL-NCE loss intuitively helps to solve this obstacle. (2-3 sentences)

2. What is the effect of increasing the number of negative examples? Explain intuitively what is happening. (2 sentences)

3. Think on other computer vision scenarios where the MIL-NCE loss could be useful and explain why. (3 sentences)