Meta-training, which is also called model agnostic machine learning (MAML), is the idea of teaching the model to create a solver to the problem in lieu of teaching the model to learn to solve a problem. MarkTechPost brought this interesting topic to our attention in their article, “New AI Research From Deepmind Explains How Few-Shot Learning (FSL) Emerges Only When The Training Data Is Distributed In Particular Ways That Are Also Observed In Natural Domains Like Language.”

The discovery that many natural data sources, including natural language, deviate from normally supervised datasets due to a few significant traits inspired this idea.

Meta-training, on the other hand, entails training a model directly on specifically designed data sequences in which item classes only recur and item-label mappings are only stable within episodes — not across episodes. Meta-training is sometimes called few shot learning as the optimized model requires a few extra training loops on test data before it can be used to predict.

Data Harmony is a fully customizable suite of software products designed to maximize precise, efficient information management and retrieval. Our suite includes tools for taxonomy and thesauri construction, machine aided indexing, database management, information retrieval and explainable artificial intelligence.

Melody K. Smith

Data Harmony is an award-winning semantic suite that
leverages explainable AI.
Learn More

Sponsored by Data Harmony, harmonizing knowledge for a better search experience.