The project:
The Transportation Security Administration (TSA)—an agency of the U.S. Department of Homeland Security (DHS)—is responsible for the safety and security of transportation systems within the United States. For the average American, the primary interaction with the TSA is during commercial air travel, as TSA runs security checkpoints at airports throughout the country.
The DHS’s Science and Technology Directorate (S&T) developed the next-generation Screening at Speed (SaS) program, which has been utilizing artificial intelligence (AI) and machine learning (ML) to create a passenger experience that is both more efficient and more secure.
The SaS program leverages the development of ML models for security checkpoints at U.S. airports. These models augment TSA officers in detecting prohibited items and people/objects of interest while reducing delays and pat-downs that frequently slow down passenger lines in current conditions.
The challenge:
The intended end result of the SaS program is a transformed air travel experience. However, the ML models driving this program must perform effectively in order for the safety of U.S. air travel to be fully maintained. All it takes is one model failure or attack to lead to a catastrophe that puts lives at risk both in the air and on the ground.
The DHS required testing and validation of these models—among others—in order to push the SaS program toward reality. Without a set process for testing and validation, it may take months or longer to identify model weaknesses, ultimately delaying the SaS project overall. The DHS needed a partner that would bring speed to model validation, as well as a repeatable layer of confidence.
CalypsoAI’s solution:
CalypsoAI partnered with the DHS to integrate its solution, VESPR Validate, into its workflow to efficiently validate and expedite the models to production. Considering the large number of models being developed for SaS, as well as the criticality and sensitivity of the mission, a repeatable testing and validation solution is essential to widespread confidence in the models. With confidence, the models can be deployed to production
CalypsoAI and the DHS have developed a systematic approach to implementing AI across these systems. These protocols will continue to be iterated and improved upon as the program matures. This will lead to a repeatable model development pipeline that expedites AI to deployment with the confidence that the models will perform exactly as intended in these highly sensitive environments.
The implementation of this process saves time in model development and mitigates uncertainty around the efficacy of the ML models in these applications, which would delay the success of the SaS program and lessen confidence in AI overall.
The future of this technology:
The DHS has stated that future operational uses of this technology could include stadiums, special events, mass transit, and more.
Conclusion:
CalypsoAI’s work with DHS has grown continuously across three phases, partnering with the SaS program to deploy the next generation of traveler experience with confidence, security, and speed.
