ML versus DL
There is three term that we use interchangeably but are subtly different.
- Artificial Intelligence, or AI
- Machine learning, or ML
- Deep learning, or DL
Artificial Intelligence is the theory of machines trying to mimic how human thinks, interact, takes decision. Artificial Intelligence is the superset as all ML or DL algorithm are AI too. However, there are algorithms in AI that are not necessarily ML or DL too. For instance we can program a computer to beat human in chess without using ML. Those usages require complexes algorithms and a lot of human intelligence to ponder huge decision tree that may arise.
Machine Learning learns pattern between structured, already "processed" data, in order to mimic how human takes decision. Machine learning is a subset of Artificial Intelligence and a superset of Deep Learning. Typical ML algorithms that are not also DL algorithm such as Random Forest or Boosted Tree. They "learn more" that statistic model (in time series, algorithm like PROPHET or ARIMA) but still require data as input.
Deep Learning learns pattern between raw data. All the recent algorithm, such as ResNet and Transformer based, are deep learning. Technically, all of them take a list of number in inputs and give a list of number as outputs. They can do a lot of different tasks.
Let looks at the differences.
| Category | "Human Thinking" needed | "Data and training ressources" needed |
|---|---|---|
| AI | A lot | None |
| ML | A medium amount (*) | A medium amount |
| DL | Almost None (*) | A LOT !!!! |
* : At least for already studied problems (text recognition, generative
content, etc). If you try to do innovation and try to apply ML or DL to "blue
ocean" fields, you will need a lot of brain power and experiments