Lifelong Learning Machines (L2M)
Abstract
The Lifelong Learning Machines (L2M) program will research and develop fundamentally new machine learning mechanisms, enabling machines that learn continuously as they operate. Current learning machines are fully configured in advance of deployment, meaning that they have difficulty accounting for in-the-field mission changes or for unexpected deviations in the data being processed. To overcome this limitation, L2M will pursue learning approaches inspired by biological systems, which continuously learn and improve their skills without losing previous knowledge. Areas of research will include network structures that improve performance by processing new data seen in the field, learn new tasks without forgetting previous tasks, and incorporate context into their understanding of the environment. These capabilities would impact a broad array of military applications that require processing and understanding data in real-time, often have limited data sets for training, and must be deployed in environments where unpredictable events may occur. The L2M program moves to Project ES-02, Beyond Scaling Sciences, in FY 2019.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2020
- Source ID
- d8c248ea0e633bc7abc06505e44c09df