New method allows AI to learn without data

The researchers explained that machine learning requires a lot of examples from the data.For example, in order to

create an AI model that allows you to recognizehorse, she needs to analyze thousands of horse images. This is what makes technology expensive and different from human training. A child often only needs to see a few examples of an object, or even one, before he can recognize it throughout his life.

The new paper suggests that AI models toocan be trained this way - scientists called this process "less than one" - when the algorithm recognizes more objects, despite the fact that the amount of data on which it was trained was small.

For example, researchers trained AI to recognizenumbers, but they did not load data about each number into the model, but did it as a single picture, taking into account that many numbers have similar styles. This allowed them to reduce the amount of data from 60 thousand images to 10. 

AI has learned to select materials for research

Researchers are now working tofind other ways to design small synthetic datasets, be it manual design or using another algorithm. However, despite these additional research challenges, the article provides a theoretical basis for further learning. “Our takeaway is that no matter what datasets you have, you can probably package them up to make the model more efficient,” they said.

In the future, researchers want to train even powerfulmodels based on small data sets. In doing so, they will draw up clear instructions for packaging data so that scientists with even little experience can use them.

Read also

Previously, the Earth and the Moon had a common magnetic shield. It saved the planet from the sun

There may be universes in black holes. We tell you about the new discovery

On day 3 of illness, most COVID-19 patients lose their sense of smell and often suffer from a runny nose