Zero-shot learning
lastmod 2023-01-10created 2022-11-114 edits
Zero-shot learning is a problem setup in machine learning where, at test time, a learner observes samples from classes that were not observed during training and must predict the class they belong to.
Abstract
Generalized zero-shot learning (GZSL) aims to train a model for classifying data samples under the condition that some output classes are unknown during supervised learning. To address this challenging task, GZSL leverages semantic information of the seen (source) and unseen (target) classes to bridge the gap between both seen and unseen classes. Since its introduction, many GZSL models have been formulated. In this review paper, we present a comprehensive review on GZSL. Firstly, we provide an overview of GZSL including the problems and challenges. Then, we introduce a hierarchical categorization for the GZSL methods and discuss the representative methods in each category. In addition, we discuss the available benchmark data sets and applications of GZSL, along with a discussion on the research gaps and directions for future investigations.
Documentation
- A Review of Generalized Zero-Shot Learning Methods
- https://arxiv.org/abs/2011.08641
About
Put simply, zero-shot means "a model performing a task it did not learn during training." Examples from generative models include the following.
- A TTS model trained to generate speech in Yoo In-na's voice using example samples to also generate speech in IU's voice
- A natural language generation model trained to write like Shakespeare writing in the style of Mark Twain
- Generating kinds of images that did not exist during training
Beyond these, there are many applications that perform tasks that were not directly trained. Enabling a model to perform a specific task even though it was never taught that task during training is called zero-shot learning.