5,337
edits
Line 1,475: | Line 1,475: | ||
Our typical example has <math>\Vert x_i \Vert^2 = O(d)</math>. | Our typical example has <math>\Vert x_i \Vert^2 = O(d)</math>. | ||
Consider <math>x^{test} = 0</math> then <math>P_{\theta}(x^{test}) > P_{\theta}(x_1)</math>. | Consider <math>x^{test} = 0</math> then <math>P_{\theta}(x^{test}) > P_{\theta}(x_1)</math>. | ||
==Domain Adaptation== | |||
So far, we have a training set <math>\{(x_i^{(train)}, y_i^{(train)})\}</math> from distribution <math>Q_{X,Y}</math>. | |||
We learn optimal parameters <math>\theta^*</math> via ERM. | |||
Then at test time, our test samples come from the same distribution <math>Q_{X,Y}</math>. | |||
However in practice, the training distribution can be different from the test distribution. | |||
The training distribution is the source domain. The test distribution is the target domain. | |||
;Examples | |||
Q may be synthetic samples and P may be real samples. | |||
Q contains samples with white background but P has samples with real backgrounds. | |||
In training: | |||
For the source domain, we have labeled samples <math>\{(x_i^S, y_i^S)\}_{i=1}^{m_S} \sim Q_{X,Y}</math>. | |||
For the target domain, we only have unlabeled samples <math>\{x_i^t\} \sim P_{X}</math>. | |||
==Misc== | ==Misc== |