The MADLab is a University Center of Excellence supported by the Air Force Office of Sponsored Research (AFOSR) and the Air Force Research Lab (AFRL). The Center is developing the next-generation of Machine Learning (ML) theory, algorithms, and applications.
ML has advanced considerably in recent years, but mostly in well-defined domains using huge amounts of human-labeled training data. Many application domains may not be so well-defined and, while data rich, they tend to be label poor. This issue has limited the impact of state-of-the-art ML theory and methods in specialized applications. Furthermore, the decision-making mechanisms of existing ML systems are often difficult to interpret, calling into question the use of such systems in mission-critical operations. Adding to these challenges are the need to fuse information from multiple sensors, adapt to dynamic environments, cope with missing or severely corrupted data, design more efficient ML hardware, and combat adversarial forces aiming to disrupt ML systems.
These issues present several grand challenges for ML that are the Center’s research foci:
- new ML theory and methods to enable the rapid training and/or adaptation of predictors using a small number of labeled training examples;
- novel ML systems that are implemented efficiently in both hardware and software, and ML solutions that are easily scalable, from cloud-based systems to low-power devices;
- ML algorithms that are interpretable as well as robust to dynamic operational conditions, missing data, and sensor failures; and
- new ML technology that is robust to adversarial attacks and data contamination.
The Center is also spearheading the development of new ML technology tailored to specialized applications, as well as training the next generation of machine learning researchers and practitioners.