Catalogue of Artificial Intelligence Techniques
Aliases: Transductive Learning
Keywords: classification, learning, transduction
Author(s): Oliver Wilson
Unlike inductive learning, where the aim is to find a general function f that maps inputs x to some output classes y that should be accurate for any future test set, transduction aims to predict classes for a specific test set. It does this by drawing conclusions directly from the training data as opposed to producing a general model and then applying it.
The technique is useful in areas where unlabeled data is relatively cheap/easy to come by and labeled data is very costly (for instance, it must be manually classified by a human). Transductive learning is closely related to semi-supervised learning. Transductive learning was first developed in the 90's by Vladimir Vapnik.
- Ran El-Yaniv and Dmitry Pechyony, Stable Transductive Learning.
- Olivier Bousquet, Transductive Learning: Motivation, Model, Algorithms, 2002.
- Thorsten Joachims, Transductive Inference for Text Classification using Support Vector Machines, 1999.