Why probabilistic ML?
- Consistent estimation: guaranteed to benefit from large data sets
- Interpretable model weights, interpretable latent space
- Sufficient statistics: fast learning from arbitrary large data sets
- Generative modelling: ready to sample new data, detect novelties, replace missing values
- Probabilistic inference: report marginal probabilities for interesting sub-sets of variables, perform maximum a posteriori prediction
- Expectation maximization: learn from missing data