Please turn your device

Machine Learning With Less Than One Example Is Now A Reality

Machine learning usually needs numerous examples in order to be effective. 

Unlike human beings that only need to see something a couple of times before they learn, machines need a ton of reps and examples of the same information in order to recognize and “learn” it. 

That reality seems to be changing. 

New research from the University of Waterloo in Ontario has opened the possibility for a new machine learning process that will only need one example in order to learn. 

The process called “less than one”-shot, or LO-shot, empowers a machine learning model trained to perform one task, to perform a related task with a single or very few new examples. To put it simply, an AI model will be able to recognize more objects than the number of examples it was originally trained on. 

This might all sound like technical “blah-blah” to you but this could very well be a turning point for the entire artificial intelligence and machine learning industry.

The research might still be in its early stages but it is essentially proposing a novel solution to a problem that was otherwise thought to be unsolvable – making AI functionable without the need of data silos.

Here’s what Tongzhou Wang, an MIT PhD student who led the earlier research on data distillation had to say about the new findings:

The paper builds upon a really novel and important goal: learning powerful models from small data sets.” 

This is a major development for both companies developing AI technologies, but most importantly, for the companies using it.

This new model is making AI more accessible to companies and industries that have thus far been stifled by the extremely demanding data requirements. 

Using less data to yield the same result can democratize the use of artificial intelligence and close the gap between conglomerates and SMBs. 

Another area that could be directly affected by the “less than one”-shot, is data privacy. 

Logic dictates that since less information would have to be extracted in order to create a fully-functional AI model, the smaller the possibility for data leaks, malicious attacks and any other malevolent digital threat. 


The research might still be in its infancy, but it’s most definitely something that has us excited. Unlocking this specific part of artificial intelligence might actually be the most interesting development in the field since…the inception of the actual field.  

We could have gone into much greater detail analyzing the Xs and Os of the paper and the technology behind “less than one”-shot, but we decided to give you a high-level overview of what it is, what it does and how it can impact interested stakeholders. 

The power and potential of AI technology has been well-established and this is a step in realizing it. Stay tuned as we will keep an eye on this technology and report on it as soon as there are more developments.