Many recent video applications---including traffic monitoring, drone analytics, autonomous driving, and virtual reality---require piecing together, combining, and operating over many related video streams. Despite the massive data volumes involved and the need to jointly reason (both spatially and temporally) about these videos, current techniques to store and manipulate such data are often limited to file systems and simple video processing frameworks that reason about a single video in isolation. We present Apperception, a new type of database management system optimized for geospatial video applications. Apperception comes with an easy to use data model to reason about multiple geospatial video data streams, and a programming interface for developers to collectively reason about the entities observed in those videos. Our demo will let users write queries over video using Apperception and retrieve (in real-time) both metadata and rendered video data. Users can also compare results and observe speedups achieved by using Apperception.

Recurrent neural networks are widely used on time series data, yet such models often ignore the underlying physical structures in such sequences. A new class of physics-based methods related to Koopman theory has been introduced, offering an alternative for processing nonlinear dynamical systems. In this work, we propose a novel Consistent Koopman Autoencoder model which, unlike the majority of existing work, leverages the forward and backward dynamics. Key to our approach is a new analysis which explores the interplay between consistent dynamics and their associated Koopman operators. Our network is directly related to the derived analysis, and its computational requirements are comparable to other baselines. We evaluate our method on a wide range of high-dimensional and short-term dependent problems, and it achieves accurate estimates for significant prediction horizons, while also being robust to noise.