SSL Review -1

小样本综述:A Survey of Zero-Shot Learning: Settings, Methods,and Applications


Classify zero-short learning according to the data utilized in model optimization

different semantic spaces adopted in existing zero-shot learning works

categorize existing zero-shot learning methods

different applications of zero-shot learning

future research directions of zero-shot learning


Resttrictions for Supervised classification

  • 1、sufficient labeled training instances are needed for each class
  • 2、the learned classifier can only classify the instances belonging to classes covered by the training data
  • 3、 it lacks the ability to deal with previously unseen classes
  • in practical applications , there may not be sufficient training instances for each class.
    there could also be situations in which the classes not covered by the training instances appear in the testing instances.

Method to deal with the problems

  • few-shot learning methods
    one-shot learing methods

    • in these methods , while learning classifiers for the classes with few instances , knowledge contained in instances of other classes is utilized .
  • open set recognition methods ,

    • when learning classifier with the training data , the fact of unseen classes existing is taken in consideration
    • the learned classifier can determine whether a testing instance belong to the unseen classes , but it cannot determine which specific unseen class the instance belongs to.
  • the cumulative learning methods
    the class-incremental learning methods

    • proposed for problems in which labeled instances belonging to some previously unseen classes progressively appear after model learning .
    • the learned classifier can be adapted with these newly available labeled instances to be able to classify classes covered by them
  • the open world recognition methods

    • follow the process of “unseen classes detection , labeled instances of unseen classes acquisition, and model adaption “
    • adapt the classifier to be able to classify previously unseen classes with the acquired labeled instances belonging to them
  • disadvantage of the methods under the above learning paradigms

    • if the testing instances belong to unseen classes that have no available labeled instances during the model learning .
    • the learned classifier cannot determine the class labels of them.
    • However , in many practical applications , we need the classifier to have the ability to determine the class labels for the instances belonging to these classes .
  • Some popular application scenarios

    • The number of target classes is large.

      • example: activity recognition
    • Target classes are rare.

      • example: fine-grained object classification , recognize flowers of different breeds.
    • Target classes change over time.

      • example: recognizing images of products belonging to a certain style and brand.
    • In some particular tasks which is expensive to obtain labeled instances.

  • For a classifier

    • it is important for it to have the ability to determine the class label of instances belonging to these classes .
    • to solve this problem ,zero-shot learning is proposed .
  • Zero-shot learning

    • aim to classify instances belong to the classes that have no labeled instances.

Overview of Zero-Shot learning

XMind: ZEN - Trial Version