107 - Multi-Modal Transformers, with Hao Tan and Mohit Bansal
NLP Highlights
00:00
Do You Have a Future Regression Task?
The future regression task is basically trying to do model distillation of the original object detector that was trained. So you're not taking the original pictles and producing the feature representations. That would be model distillation. You're doing something much more like autoin coating, which is like, i take my features, i compress them or re jigger them in some way, and then try to predict them again. We generally use five tete set. And its group in two categories. The first wine ve cue the second lines, cicure, and the third wines feccu. In tota, five different dateses.
Transcript
Play full episode
