Existing semi-supervised learning algorithms focus on vectorial data given in Euclidean space. But many real life data are non-metric, given as (dis-)similarities which are not widely addressed. We propose a conformal prototype-based classifier for dissimilarity data to semi-supervised tasks. A ‘secure region’ of unlabeled data is identified to improve the trained model based on labeled data and to adapt the model complexity. The new approach (i) can directly deal with arbitrary symmetric dissimilarity matrices, (ii) offers intuitive classification by sparse prototypes, (iii) adapts the model complexity. Experiments confirm the effectiveness of our approach in comparison to state-of-the-art methods.
- Semi-supervised learning
- Proximity data
- Dissimilarity data
- Conformal prediction
- Generalized Learning Vector Quantization