Algorithms
|
Many different (SVM, DT, kNN, …) |
Defined by architecture (RNN, GAN, LSTM, ...) |
Data size
|
Can work well with smaller inputs |
Requires large amount of data |
Performance
|
Typically extremely fast |
Computational complexity depends on the architecture |
Features
|
Hand-crafted |
Can be learned |
Preprocessing
|
Significant effort |
Can be trained on raw data |
Fine tuning
|
Setting the algorithm parameters |
Can be performed automatically during training |
Complexity
|
Typical simple mathematical models |
Depends on the architecture (highly flexible) |
Transparency
|
Typically transparent |
Hard to transparently show decision making |
Explainability
|
Typically explainable |
Hard to show the reasoning process |