With the public availability of medium-resolution, high revisit-rate satellite imagery with worldwide coverage, such as Landsat and Sentinel-2, it is now feasible to perform automated, world-scale detection of interesting anthropogenic events. Challenges include disambiguating events from other changes in the scene and sensor, limited image resolution relative to the events to be detected, and difficulty in acquiring large amounts of labeled training data. An effective approach to this problem needs to leverage the latest advances in machine learning, remote sensing, and cloud computing. The IARPA SMART program recently evaluated the performance of state-of-art approaches in real-world test scenarios for the specific case of heavy construction events. Here we describe the approach used by the Systems and Technology Research team to achieve top-performing broad area search (BAS) accuracy in Phase I of SMART, which leverages recent advances in multispectral material classification, deep semantic segmentation, and probabilistic reasoning.

The loss landscapes of deep neural networks are not well understood due to their high nonconvexity. Empirically, the local minima of these loss functions can be connected by a learned curve in model space, along which the loss remains nearly constant; a feature known as mode connectivity. Yet, current curve finding algorithms do not consider the influence of symmetry in the loss surface created by model weight permutations. We propose a more general framework to investigate the effect of symmetry on landscape connectivity by accounting for the weight permutations of the networks being connected. To approximate the optimal permutation, we introduce an inexpensive heuristic referred to as neuron alignment. Neuron alignment promotes similarity between the distribution of intermediate activations of a model along the curve with that of the endpoint models. We provide theoretical analysis establishing the benefit of alignment to mode connectivity based on this simple heuristic. We empirically verify that the permutation given by alignment is locally optimal via a proximal alternating minimization scheme. Empirically, optimizing the weight permutation is critical for efficiently learning a simple, planar, low-loss curve between networks that successfully generalizes. Our alignment method can significantly alleviate the recently identified robust loss barrier on the path connecting two adversarial robust models and find more robust and accurate models on the path.