IARPA releases Thor BAA
On June 15, the Intelligence Advanced Research Projects Activity released a broad agency announcement for the Thor program (Solicitation Number: IARPA-BAA-16-04). IARPA will accept questions until July 12, 2016. Proposal Due Date for Initial Round of Selections: 5:00 pm Eastern Time August 15, 2016.
The Thor program seeks to develop biometric presentation attack (PA) detection technologies to detect when someone is attempting to disguise their biometric identity to circumvent biometric security systems. Proposed approaches must be capable of detecting known and unknown presentation attacks. Biometric modalities of interest are face, finger, and iris.
The goal of this program is to utilize Presentation Attack Detection (PAD) to identify known and unknown Presentation Attacks (PA) in a biometric collection system. A biometric PA, also commonly referred to as biometric spoofing, is a method which inhibits the intended operation of a biometric capture system, interfering with the recording of the true sample/identity, ultimately preventing the subject from being correctly identified. Typical PAs utilize a prosthetic to conceal the biometric signature or present an alternative biometric signature.
Existing technology in use primarily relies upon a human security presence to ensure the integrity of the process and that a PA is not being utilized. There are some minimal PAD technologies in use, primarily focused on detecting a specific subset of known PAs. It is anticipated that the use of biometric collection systems will continue to increase. As we become increasingly reliant upon this technology to adjudicate identity, it is important that the technology cannot be easily deceived utilizing a PA. Additionally, reliance upon a human in the loop is cost prohibitive for many applications. Existing PAD approaches focus on methods such as Liveness Detection, Intrinsic Sample Properties, or Artificial Indicators as shown in Table 1. Current sensor hardware captures limited information pertinent to PAD with no intelligence to identify zero-day unknown PAs. A need exists to capture more robust information from a biometric sample to identify, or measure likelihood of, PAs. There needs to be an ‘intelligent’ approach that can identify unknown presentation attacks based on knowledge of what a true sample should look like (e.g., normalcy modeling for anomaly detection).
The program is anticipated to be divided into three phases. Phase 1 will last for a period of 18 months and will focus on the ability to detect known PAs. Phase 2 will be 18 months and will focus on the ability to detect unknown PAs. Phase 3 will be 12 months and will focus on operationally relevant performance requirements. Following the conclusion of Phases 1 and 2, respectively, down-selection is possible for a variety of reasons including but not limited to underperforming PAD modalities or proposals.
Full information is available here.