Artificial intelligence software from MIT and Georgetown University Medical Center uses prior learning to more efficiently understand visual concepts.


  • Increased speed
  • Increased efficiency
  • Dynamic


  • AI/Machine Learning

UN Sustainable Development Goals Addressed

  • Goal 9: Industry Innovation & Infrastructure

The Challenge

Humans have the ability to quickly and accurately learn new visual concepts without a lot of data, oftentimes from just a single example. Even toddlers can learn to quickly recognize zebras and distinguish them from cats, horses, and giraffes. This is because the brain is able to simplify learning by using previously learned representations to inform new information. Computers, however, oftentimes need to see an example hundreds or even thousands of times to learn what it is and distinguish it from a similar object.

Innovation Details

The AI (artificial intelligence) software allows computers to learn new visual concepts with only a small number of examples, similar to how the brain learns. It does this by identifying relationships between visual categories to differentiate visual data, rather than just identifying single characteristics such as color and shape. For example, rather than identifying the basic visual features of a platypus, it stores more high-level concepts, like how a platypus looks a bit like a duck, a beaver, and a sea otter. The computer then stores this information in a databank that can be easily referenced and associated with new, incoming information. Altogether, this increases the efficiency and speed of visual learning in computers.


Biological Model

As the brain learns, it builds networks of connected neurons to store and share information including, size, color, and texture of an object. One area of the brain, the anterior temporal lobe, is thought to contain complex neural hierarchies that code for additional “abstract” concept representations that go beyond the basic characteristics of color and shape. This ‘database’ of information is constantly being updated and referenced whenever we learn a new task or object. This allows humans to learn new information much more quickly and easily.