Stop asking your neural network to "Select one of the following ..." and start asking it to "Write a paragraph explaining ...". When you give your model more information in the training output, you are teaching your model from an essay instead of multiple choice. This increase of information can lead to an increase in accuracy and is why you should consider giving your model multiple "tasks" or problems to solve with prediction.

In this presentation, I will discuss how Hudl is using multi-task models to clip sports videos into plays and also how we determine the formation of an offense in American football. After we talk about what multi-task models are and how they can be used, I will share the some of the challenges we have run into that differ from a traditional model output.