The availability of moderate and high resolution satellite imagery collections and advancements in large scale cloud computing have created opportunities to develop globally consistent and near real time (NRT) land use and land cover (LULC) classification products. Such products have in turn facilitated the monitoring of land use and land cover change (LULCC) at a much larger scale than before. This is essential in understanding and analysing the competition for space in the context of a rapidly growing population, increasing standard of living, and a strained ecosystem. Several studies have pointed towards inaccuracies and spatial and typology bias in LULC classification products. In this paper we investigate the feasibility of utilizing publicly available, pretrained models for LULCC analysis. We describe a framework for large scale analysis that allows incorporating diverse sources of data and we discuss a case study of quantifying LULCC driven by green energy infrastructure expansion. We introduce the concept of Rapid Change Sequences which can be used to improve classification accuracy. Lastly, we propose a method to produce vector embeddings of the LULC change graph, which is the first tool to allow cross country comparisons of land use change dynamics.

To support the replication of ?Human Object Interaction Detection via Multi-level Conditioned Network", which was presented at ICMR'20, this companion paper provides the details of the artifacts. Human Object Interaction Detection (HOID) aims to recognize fine-grained object-specific human actions, which demands the capabilities of both visual perception and reasoning. In this paper, we explain the file structure of the source code and publish the details of our experiments settings. We also provide a program for component analysis to assist other researchers with experiments on alternative models that are not included in our experiments. Moreover, we provide a demo program for facilitating the use of our model.

We present PyTorch Geometric Temporal, a deep learning framework combining state-of-the-art machine learning algorithms for neural spatiotemporal signal processing. The main goal of the library is to make temporal geometric deep learning available for researchers and machine learning practitioners in a unified easy-to-use framework. PyTorch Geometric Temporal was created with foundations on existing libraries in the PyTorch eco-system, streamlined neural network layer definitions, temporal snapshot generators for batching, and integrated benchmark datasets. These features are illustrated with a tutorial-like case study. Experiments demonstrate the predictive performance of the models implemented in the library on real-world problems such as epidemiological forecasting, ride-hail demand prediction, and web traffic management. Our sensitivity analysis of runtime shows that the framework can potentially operate on web-scale datasets with rich temporal features and spatial structure.

We consider the problem of finding large scale mobility patterns. A common challenge in mobility tracking systems is that large quantity of data is spread out spatially and temporally across many tracking sensors. We thus devise a spatial sampling and information exchange protocol that provides probabilistic guarantees on detecting prominent patterns. For this purpose, we define a general notion of significant popular paths that can capture many different types of motion. We design a summary sketch for the data at each tracking node, which can be updated efficiently, and then aggregated across devices to reconstruct the prominent paths in the global data. The algorithm is scalable, even with large number of mobile targets. It uses a hierarchic query system that automatically prioritizes important trajectories - those that are long and popular. We show further that this scheme can in fact give good results by sampling relatively few sensors and targets, and works for streaming spatial data. We prove differential privacy guarantees for the randomized algorithm. Extensive experiments on real GPS data show that the method is efficient and accurate, and is useful in predicting motion of travelers even with small samples.

Searching for similar GPS trajectories is a fundamental problem that faces challenges of large data volume and intrinsic complexity of trajectory comparison. In this paper, we present a suite of sketches for trajectory data that drastically reduce the computation costs associated with near neighbor search, distance estimation, clustering and classification, and subtrajectory detection. Apart from summarizing the dataset, our sketches have two uses. First, we obtain simple provable locality sensitive hash families for both the Hausdorff and Fréchet distance measures, useful in near neighbour queries. Second, we build a data structure called MRTS (Multi Resolution Trajectory Sketch), which contains sketches of varying degrees of detail. The MRTS is a user-friendly, compact representation of the dataset that allows to efficiently answer various other types of queries. Moreover, MRTS can be used in a dynamic setting with fast insertions of trajectories into the database.