اوکی.منظور شما تبدیل این ساختار به ساختار با ناظر هست. خوب این روشهای مختلفی داره و مقالات مختلف. دو تا راه مشهور هست که من عینا از متن مقاله براتون میارم و خود مقاله رو هم براتون اخر نظر میزارم که بتونید از مراجع و متن خود مقاله استفاده کنید.
Two ways of transforming the unsupervised models into classifiers have been
used: either use the pre-trained model as a feature extractor, and use the concatenation of all hidden
representations as inputs to a classifier (classically, an SVM) [11, 15], or use pre-trained weights to
initialize a deep feed-forward multi-layer neural network, that will then be fine-tuned by gradient
descent on a supervised training criterion [3, 4, 5].
1
Other types of deep learning have been explored, that combine supervised and unsupervised updates
at the same time, for instance [16]. Instead, we focus here on algorithms with the above two-
phase distinction, and we wish to compare the flavors that perform supervised fine-tuning of the
representations with those that do not.
چیزی که در نظر قبل بهتون پیشنهاد دادم روش دوم این مقاله هست یعنی وزن هاتون رو در بیارید و بریزید داخل یه MLP و بعد با استفاده از Backpropagation اموزش بدید.
رفرنس:
Lamblin, Pascal, and Yoshua Bengio. "Important gains from supervised fine-tuning of deep architectures on large labeled sets." NIPS* 2010 Deep Learning and Unsupervised Feature Learning Workshop. 2010.