There is a growing need in a wide range of industries to leverage AI to assist in creating operational plans. In this research, we present an AI approach for learning operational plan construction by concurrently learning Bayesian agents and deep reinforcement learning (DRL) agents that explore orthogonal parts of the decision-space. Within our concurrent learning, we use a hierarchical Bayesian program learning (HBPL) structure which provides a mechanism to incorporate a degree of structure to the learning process by leveraging knowledge from expert operational planners. We demonstrate that our concurrent learning approach out-performs a state-of-the-art full DRL approach in both speed and operational plan quality across a range of operational planning settings with increasing complexity of the decision-space. Moreover, we demonstrate that the gains in speed and plan utility from concurrent learning increase over the full DRL approach as decision-space complexity grows within the simulation environment. This work is applied to a specific setting, operational planning for air command and control; however, it provides a structure that is generalizable to a wide range of operational planning settings across industries. Additionally, we introduce a new architecture for HBPL that enables program learning within a reinforcement learning setting, providing an updated structure for Bayesian reinforcement learning.

Seabed characterization has utility for numerous applications that seek to explore and interact with the seafloor, ranging from coastal habitat monitoring and subbottom profiling to man-made object detection. In this paper, we characterize seabeds based on the texture patterns within synthetic aperture SONAR (SAS) images constructed from high-frequency side-scan sonar. Features are measured from the SAS images (e.g., lacunarity, an established texture feature coding method, and a circularly shifted histogram of oriented gradients). Based on these SAS image features, we perform unsupervised clustering with a hierarchical Bayesian model, which creates categories of seabed textures. Our clustering algorithm is a new variant of the hierarchical Dirichlet process that is both adaptive to changes in seabeds and processes batches of SAS imagery in an online fashion to learn new seabed types as they are encountered. This allows observations to be clustered, as each batch is processed rather than only after all data have been collected. The model's performance of seabed characterization by SAS image texture is demonstrated in the overall range and internal consistency of textures specific to each learned cluster with data across a variety of sites.

We present a Bayesian approach to unsupervised clustering of activity within video imagery. Vehicles and pedestrians are tracked within the video imagery and their collective activity in each time frame is measured and categorized using a natural extension of the dynamic latent Dirichlet allocation model. Our extension involves use of multiple types of simultaneously observed features from multiple classes of objects within the video imagery. Within the prior for the model these features are treated as independent, and modeled as draws from a variety of appropriate distribution types. By including multiple features, the model generates a richer set of activities; we quantitatively show that this yields better predictions of physical attributes within the scene, relative to currently available models that use only the single best feature. We show this by comparing model prediction of traffic light states within a busy intersection, which we ground-truth manually within the video imagery.

This paper describes an effective process for automated detection and classification of frequency-modulated sounds from birds, crickets, and frogs that have a narrow short-time frequency bandwidth. An algorithm is provided for extracting these signals from background noise using a frequency band threshold filter on spectrograms. Feature vectors are introduced and demonstrated to accurately model the resultant bioacoustic signals with hidden Markov models. Additionally, sequences of sounds are successfully modeled with composite hidden Markov models, allowing for a wider range of automated species recognition.