Zero shot learning

Similar to unsupervised learning, zero shot learning (ZSL) is a method of training an AI model to recognize and categorize objects or ideas, without having previously seen examples of those things or ideas.

Both approaches use a form un "unlabeled" data — so, the system isn't told what it is looking at, but it "self-supervises" its learning by, for instance, recognizing and comparing patterns in images or reams of text.

But, according to IBM's definition of ZSL — and this is where it gets fascinating — a ZSL can analyze "a massive corpus of text that may contain incidental references to or knowledge about unseen data classes. Without labeled examples to draw upon, ZSL methods all rely on the use of such auxiliary knowledge to make predictions."

Researchers at Oxford University in the UK have described ZSL as "inspired by the way human beings are able to identify a new object by just reading a description of it, leveraging similarities between the description of the new object and previously learned concepts." (za/fs)

 

Sources:

What is zero-shot learning? (IBM) https://www.ibm.com/topics/zero-shot-learning (accessed May 15, 2024)

An embarrassingly simple approach to zero-shot learning (Paper by Bernardino Romera-Paredes and Philip H. S. Torr) https://proceedings.mlr.press/v37/romera-paredes15.pdf (accessed May 15, 2024)

Read the rest of DW's AI Lexicon:

A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z

We're keen to hear your feedback. Suggest an entry by sending us a comment. And let us know if you feel we have missed something, got it wrong, and tell us whether our AI Lexicon has helped you understand the technology better.

Written and edited by: Zulfikar Abbany (za), Fred Schwaller (fs)