What is unsupervised DL?
Unless you are lucky, the chances are that most of the data that comes to you is unlabeled, whether it is images on the web, text from a document, gene expression data from NGS experiments, and so on. Even if they come labeled, they are not clean and perfect datasets. This is where UL algorithms are useful. In UL, the algorithm is presented with the training datasets without any label, which means these datasets don’t have a particular outcome or specific instructions on what to do with them. The job of the UL model is to automatically extract features from unlabeled datasets and use those features to find hidden patterns. The unsupervised models first try to extract simple features from the data, then stitch them together to form more advanced features, and finally, come up with an outcome. Unlike SL models, these models don’t have a ground truth to evaluate the performance of the models using metrics such as accuracy, mean squared error (MSE...
