Humans are incredible learners—able to rapidly assimilate new information on the fly to robustly learn new concepts and skills from limited experiences. In contrast, current Machine Learning (ML) systems can exceed the scale at which humans learn but are much less data efficient and are difficult and costly to update in the face of new data. This research program aims to identify key characteristics of human learning that are not yet realized within artificial systems, and then explore how to design systems with these capabilities.
Research in this area has the potential to produce systems that:
- learn more efficiently, yielding a smaller energy and carbon footprint;
- learn from less data, translating into better performance in situations where data is difficult to collect or proprietary;
- update on the fly, able to adapt to users in realtime in response to their inputs;
- learn continually, staying abreast of a constant stream of data without forgetting previous capabilities; and
- many others.
Interested? Get Involved!
Consider participating in one of the two upcoming events sponsored by AAAI:
|AAAI-24 Symposium on Human-Like Learning
|AAAI-24 Tutorial on Probabilistic Concept Formation