PIERRE JOHN M LLC — Department of Defense SBIR Phase II: OSD182-006

Description

PhysicsAI proposes to develop a robust framework for using deep reinforcement learning (DeepRL) to create optimal training data from simulated imagery in order to train machine learning models with improved accuracy, increased robustness, and which support few/zero shot detection for rare objects. Specifically we plan to extend our DeepRL sim2real prototype developed in Phase I by (1) adding the ability to use Generative Adversarial Networks and Auto Encoders/Decoders to automatically create photorealistic synthetic backgrounds; (2) scaling up our platform to support parallel training of multiple DeepRL agents and creation of larger training sets to boost detection accuracy; (3) testing and fully characterizing the performance of the framework with respect to both limited training data and generalizability to detectors to new background environments a commercial high-resolution satellite imagery dataset.