Machine Learning = Looking for a function
- Function with unknown
:the unknown parameters
- Define loss function
(
:sum over training examples in one task)
- Optimization
Meta Learning = Learn to learn
- What is learnable in a learning algorithm? Net Architecture, Initial Parameters, Learning Rate, ...
:learnable components
- Define loss function (
) for learning algorithm
the smaller the better, N is the number of the training tasks (
:sum over testing examples in one task;
:sum over training tasks)
- Find
that can minimize
, then have a learned "learning algorithm"