This document summarizes a research paper that applies neural architecture search (NAS) to feature pyramid networks (FPN) for object detection. Specifically, it searches for an optimal FPN structure by defining a search space that covers all possible cross-scale connections in FPN. The NAS process involves a controller RNN that samples "merging cells" to build the FPN, with the final outputs of the best cells forming the feature pyramid. The resulting NAS-FPN architecture achieves state-of-the-art object detection accuracy comparable to human-designed models but with faster inference speeds, demonstrating the effectiveness of the automated design approach.