- Deep Learning’s ‘Holy Grail’ is unsupervised learning
- Being able to create general systems with little training data
- However many Deep Learning systems use large supervised datasets
- The datasets have corresponding labels (metadata)
- However training can be improved with Backpropagation