Exploring unsupervised deep learning applications
Today, practitioners have been able to leverage unsupervised deep learning to tap into their unlabeled data to achieve either one of the following use cases. These have been put in descending order in terms of their impact and usefulness:
- Creating pretrained network weights for downstream tasks
- Creating general representations that can be used as-is in downstream supervised tasks by predictive supervised models
- Achieving one-shot and zero-shot learning
- Performing dimensionality reduction
- Detect anomalies in external data
- Clustering the provided training data into groups
To start, note that pure clustering is still a core application of unsupervised learning in general, but not for deep learning. Clustering is where unlabeled data is grouped into multiple arbitrary clusters or classes. This will be useful in use cases such as customer segmentation for targeted responses, or topic modeling to figure...
