Many layers learn their own features from raw data: edges, then shapes, then objects.
- Raw inputFeed pixels, audio or text directly, no hand-made features.
- Deep layersEarly layers find simple patterns, later ones combine them.
- Lots of dataDeep nets need many examples to generalise.
- GPUsMatrix maths runs in parallel on GPUs.
Explain 1/5
Source: Deep Learning (Goodfellow, Bengio, Courville)