Clips from Andrej Karpathy's talk at ICML (June 2019). I think multi-task learning is one of the most important (and understudied) subfields of machine learning. Most real world problems are multi-task. I especially find the discussion on team workflow fascinating (see
18:25). I've been thinking and working on this topic a lot lately, and will probably give a lecture on it. Here's the outline:
0:00 - Sensors
0:29 - Single-task learning challenges
4:35 - Multi-task neural network architecture
11:50 - Loss function considerations
14:34 - Training dynamics
18:25 - Team workflow
Full talk:
https://slideslive.com/38917690/multitask-learning-in-the-wilderness