The thinking has been that the country or company with the most data will win the AI race. The most popular AI algorithms have been shown to increase in accuracy and capability as the data sets that they are trained on increases.
But alternatives to massive machine learning algorithms are beginning to be explored. Two promising AI techniques are ‘few-shot’ and ‘one-shot’ AI. These algorithms are able to train on a particular task and then use the information they’ve learned from that task and then apply it to other similar tasks or new examples. This kind of approach allows AI to be applied to problems that are similar to things that are already known.
Klamer Schutte, researcher for Intelligent Imaging, said that “learning from small and limited data sets technology allows us to leverage the benefits of current Artificial Intelligence developments without needing unaffordable large annotated effort.”
Ryan Khurana, a researcher on the Montreal AI Ethics Institute, said that “most significantly, ‘less than one’-shot learning would radically reduce data requirements for getting a functioning model built. This might make AI extra accessible to corporations and industries which have up to now been hampered by the sphere’s information necessities. It might additionally enhance information privacy, as a result of much less data must be obtained in order to train the model.”