Machine learning approaches for ray-based ocean acoustic tomography.

Underwater sound propagation is primarily driven by a nonlinear forward model relating variability of the ocean sound speed profile (SSP) to the acoustic observations (e.g., eigenray arrival times). Ocean acoustic tomography (OAT) methods aim at reconstructing SSP variations (with respect to a reference environment) from changes of the acoustic measurements between multiple source-receiver pairs. This article investigates the performance of three different OAT methods: (1) model-based methods (i.e., classical ray-based OAT using a linearized forward model), (2) data-driven methods (such as deep learning) to directly learn the inverse model, and (3) a hybrid solution [i.e., the neural adjoint (NA) method], which combines deep learning of the forward model with a standard recursive optimization to estimate SSPs. Additionally, synthetic SSPs were generated to augment the variability of the training set. These methods were tested with modeled ray arrivals calculated for a downward refracting environment with mild fluctuations of the thermocline. Idealized towed and fixed source configurations are considered. Results indicate that merging data-driven and model-based methods can benefit OAT predictions depending on the selected sensing configurations and actual ray coverage of the water column. But ultimately, the robustness of OAT predictions depends on the dynamics of the SSP variations.

Unravelled multilevel transformation networks for predicting sparsely observed spatio-temporal dynamics.

In this paper, we address the problem of predicting complex, nonlinear spatio-temporal dynamics when available data are recorded at irregularly spaced sparse spatial locations. Most of the existing deep learning models for modelling spatio-temporal dynamics are either designed for data in a regular grid or struggle to uncover the spatial relations from sparse and irregularly spaced data sites. We propose a deep learning model that learns to predict unknown spatio-temporal dynamics using data from sparsely-distributed data sites. We base our approach on the radial basis function (RBF) collocation method which is often used for meshfree solution of partial differential equations. The RBF framework allows us to unravel the observed spatio-temporal function and learn the spatial interactions among data sites on the RBF-space. The learned spatial features are then used to compose multilevel transformations of the raw observations and predict its evolution in future time steps. We demonstrate the advantage of our approach using both synthetic and real-world climate data. This article is part of the theme issue 'Data-driven prediction in dynamical systems'.

Physics-incorporated convolutional recurrent neural networks for source identification and forecasting of dynamical systems.

Spatio-temporal dynamics of physical processes are generally modeled using partial differential equations (PDEs). Though the core dynamics follows some principles of physics, real-world physical processes are often driven by unknown external sources. In such cases, developing a purely analytical model becomes very difficult and data-driven modeling can be of assistance. In this paper, we present a hybrid framework combining physics-based numerical models with deep learning for source identification and forecasting of spatio-temporal dynamical systems with unobservable time-varying external sources. We formulate our model PhICNet as a convolutional recurrent neural network (RNN) which is end-to-end trainable for spatio-temporal evolution prediction of dynamical systems and learns the source behavior as an internal state of the RNN. Experimental results show that the proposed model can forecast the dynamics for a relatively long time and identify the sources as well.

A Task-Driven Feedback Imager with Uncertainty Driven Hybrid Control.

Deep Neural Network (DNN) systems tend to produce overconfident or uncalibrated outputs. This poses problems for active sensor systems that have a DNN module as the main feedback controller. In this paper, we study a closed-loop feedback smart camera from the lens of uncertainty estimation. The uncertainty of the task output is used to characterize and facilitate the feedback operation. The DNN uncertainty in the feedback system is estimated and characterized using both sampling and non-sampling based methods. In addition, we propose a closed-loop control that incorporates uncertainty information when providing feedback. We show two modes of control, one that prioritizes false positives and one that prioritizes false negatives, and a hybrid approach combining the two. We apply the uncertainty-driven control to the tasks of object detection, object tracking, and action detection. The hybrid system improves object detection and tracking accuracy on the CAMEL dataset by 1.1% each respectively. For the action detection task, the hybrid approach improves accuracy by 1.4%.