Learning with Less Labels (LwLL)
Abstract
The Learning with Less Labels (LwLL) program, addressing a key issue encountered by the Data-Driven Discovery of Models program (budgeted in PE 0602702E, Project TT-13), will develop technology to greatly reduce the amount of labeled data required to train machine learning (ML) systems. In supervised ML, the system learns by example to recognize things, such as objects in images or speech. Humans provide these examples to ML systems during their training in the form of labeled data. With enough labeled data on which to train ML systems, it is generally possible to build useful models, but training accurate models currently requires large amounts of labeled data that can be costly to obtain. LwLL will address this problem by creating ML algorithms that learn and adapt more efficiently than current ML approaches, and by formally deriving the limits of machine learning and adaptation. LwLL-based ML systems will be easier to train and use in variable, unpredictable, real-world environments.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2020
- Source ID
- 70781ef8641d1b5888fe3ee98586b119