AFRL posts machine learning RFI
On May 18, the Air Force Research Laboratory issued a request for information for Robust and Secure Machine Learning (ML) (Solicitation Number: RFI-AFRL-RIK-18-04). RFI abstracts are due by June 30.
The Air Force Research Laboratory, Information Directorate (AFRL/RI) is seeking information to better understand existing vendor offerings and the landscape of research and development (R&D) towards robust and secure machine learning techniques.
The Air Force is investigating robust and secure machine learning techniques to determine what algorithms, methods, and techniques will be necessary to ensure efficient and effective performance of current and future machine learning systems.
Technical Challenge: Current state-of-the-art machine learning algorithms have been shown to be vulnerable to adversarial attacks which can cause decreased classification confidence or misclassification. Such attacks have been shown to be effective in both the “laboratory” setting, using digital manipulation of public datasets , and in the real-world, using physical manipulations of objects . Defensive techniques have been proposed to resist these attacks, but are often quickly defeated by new adversarial methods.
A greater understanding of ML model architectures is necessary to understand the root cause of these vulnerabilities, and inform intelligent design tradeoffs between efficiency and robustness. Therefore, we expect that solutions will require deep consideration of model training and inferencing architectures to enable maximum efficiency while retaining model robustness to adversarial inputs. Furthermore, we would like to evaluate the performances on relevant AF datasets across the domain of AFRL/RI research, to include video/image classification, communications, and cyber operations.
Full information is available here.