Yi Ding

Assistant Professor at UT Dallas

Cluster-then-Label

Clustering methods (also sometimes referred to as unsupervised learning) group input data points into subsets of similar items. Clustering methods do not need any labeled data, and are thus useful in uncovering unsuspected pat- tern/structure in the data. See

Self-training

According to [1], self-training is a wrapper-algorithm that repeatedly uses a supervised learning method in the following manner. A supervised classifier is first trained with a small amount of labeled data. The classifier is then used to classify the un-labeled data. In each iteration, a part of the unlabeled data is labeled according to a current decision function. Typically, the most confident predictions are added to the labeled train-ing set. The classifier is then re-trained and the self-training procedure is repeated.

Co-training

In co-training, we use two classifiers in parallel to improve predictions. The two classifiers use different features to obtain different perspectives and uncover different correlation between input features and the label data. Each data instance is classified by the two classifiers, and the result with higher confidence is used to retrain and update both classifiers.

Active Learning

According to [1], active learning aims at detecting the most informative unlabeled samples and queries a user to label them.

Ref

[1] Stikic, M., Van Laerhoven, K., & Schiele, B. (2008, September). Exploring semi-supervised and active learning for activity recognition. In Wearable computers, 2008. ISWC 2008. 12th IEEE international symposium on (pp. 81-88). IEEE.

[2] Radu, Valentin, Panagiota Katsikouli, Rik Sarkar, and Mahesh K. Marina. “A semi-supervised learning approach for robust indoor-outdoor detection with smartphones.” In ACM SenSys, pp. 280-294. 2014.