References: 1. A practical guide to training restricted Boltzmann machines2. Classification Using discriminative restricted Boltzmann machines
In the hot research of deep learning, RBM (limiting the Boltzmann Machine) is the most important cornerstone. In the most critical pre-training process of deep learning, RBM is trained as a generation model. The advantage that RBM can fit data well is the key to its wide application in deep models. If RBM can be applied to the discriminative model, it should have a good effect. This article mainly introduces the use of RBM as a discriminant model. Simply put, it is used as a non-linear classifier. According to section 16 in Article 1, three methods for determining RBM are introduced:
① Use the RBM hidden layer output as the feature input of other standard classifiers (such as DBN); ② train one RBM for each class, and then train a softmax classifier for decision making; ③ use feature vectors and tags as the visual layer input of RBM, and train RBM as the feature vector and tag generation model. ② is the main content of this article, which is not clearly stated, detailed descriptions are as follows:
Yes yes yes yes
From Weizhi note (wiz)
