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.

Blind deconvolution of sources of opportunity in ocean waveguides using bilinear channel models.

A general blind deconvolution algorithmic framework is developed for sources of opportunity (e.g., ships at known locations) in an ocean waveguide. Here, both channel impulse responses (CIRs) and unknown source signals need to be simultaneously estimated from only the recorded signals on a receiver array using blind deconvolution, which is generally an ill-posed problem without any a priori information or additional assumptions about the underlying structure of the CIRs. By exploiting the typical ray-like arrival-time structure of the CIRs between a surface source and the elements of a vertical line array (VLA) in ocean waveguides, a principle component analysis technique is applied to build a bilinear parametric model linking the amplitudes and arrival-times of the CIRs across all channels for a variety of admissible ocean environments. The bilinear channel representation further reduces the dimension of the channel parametric model compared to linear models. A truncated power interaction deconvolution algorithm is then developed by applying the bilinear channel model to the traditional subspace deconvolution method. Numerical and experimental results demonstrate the robustness of this blind deconvolution methodology.

Multichannel myopic deconvolution in underwater acoustic channels via low-rank recovery.

This paper presents a technique for solving the multichannel blind deconvolution problem. The authors observe the convolution of a single (unknown) source with K different (unknown) channel responses; from these channel outputs, the authors want to estimate both the source and the channel responses. The authors show how this classical signal processing problem can be viewed as solving a system of bilinear equations, and in turn can be recast as recovering a rank-1 matrix from a set of linear observations. Results of prior studies in the area of low-rank matrix recovery have identified effective convex relaxations for problems of this type and efficient, scalable heuristic solvers that enable these techniques to work with thousands of unknown variables. The authors show how a priori information about the channels can be used to build a linear model for the channels, which in turn makes solving these systems of equations well-posed. This study demonstrates the robustness of this methodology to measurement noises and parametrization errors of the channel impulse responses with several stylized and shallow water acoustic channel simulations. The performance of this methodology is also verified experimentally using shipping noise recorded on short bottom-mounted vertical line arrays.

Round-robin multiple-source localization.

This paper introduces a round-robin approach for multi-source localization based on matched-field processing. Each new source location is estimated from the ambiguity function after nulling from the data vector the current source location estimates using a robust projection matrix. This projection matrix effectively minimizes mean-square energy near current source location estimates subject to a rank constraint that prevents excessive interference with sources outside of these neighborhoods. Numerical simulations are presented for multiple sources transmitting through a fixed (and presumed known) generic Pekeris ocean waveguide in the single-frequency and broadband-coherent cases that illustrate the performance of the proposed approach which compares favorably against other previously published approaches. Furthermore, the efficacy with which randomized back-propagations may also be incorporated for computational advantage is also presented.

Passive ultrasonics using sub-Nyquist sampling of high-frequency thermal-mechanical noise.

Monolithic integration of capacitive micromachined ultrasonic transducer arrays with low noise complementary metal oxide semiconductor electronics minimizes interconnect parasitics thus allowing the measurement of thermal-mechanical (TM) noise. This enables passive ultrasonics based on cross-correlations of diffuse TM noise to extract coherent ultrasonic waves propagating between receivers. However, synchronous recording of high-frequency TM noise puts stringent requirements on the analog to digital converter's sampling rate. To alleviate this restriction, high-frequency TM noise cross-correlations (12-25 MHz) were estimated instead using compressed measurements of TM noise which could be digitized at a sampling frequency lower than the Nyquist frequency.

Motion-adaptive spatio-temporal regularization for accelerated dynamic MRI.

Accelerated magnetic resonance imaging techniques reduce signal acquisition time by undersampling k-space. A fundamental problem in accelerated magnetic resonance imaging is the recovery of quality images from undersampled k-space data. Current state-of-the-art recovery algorithms exploit the spatial and temporal structures in underlying images to improve the reconstruction quality. In recent years, compressed sensing theory has helped formulate mathematical principles and conditions that ensure recovery of (structured) sparse signals from undersampled, incoherent measurements. In this article, a new recovery algorithm, motion-adaptive spatio-temporal regularization, is presented that uses spatial and temporal structured sparsity of MR images in the compressed sensing framework to recover dynamic MR images from highly undersampled k-space data. In contrast to existing algorithms, our proposed algorithm models temporal sparsity using motion-adaptive linear transformations between neighboring images. The efficiency of motion-adaptive spatio-temporal regularization is demonstrated with experiments on cardiac magnetic resonance imaging for a range of reduction factors. Results are also compared with k-t FOCUSS with motion estimation and compensation-another recently proposed recovery algorithm for dynamic magnetic resonance imaging. .

Compressive matched-field processing.

Source localization by matched-field processing (MFP) generally involves solving a number of computationally intensive partial differential equations. This paper introduces a technique that mitigates this computational workload by "compressing" these computations. Drawing on key concepts from the recently developed field of compressed sensing, it shows how a low-dimensional proxy for the Green's function can be constructed by backpropagating a small set of random receiver vectors. Then the source can be located by performing a number of "short" correlations between this proxy and the projection of the recorded acoustic data in the compressed space. Numerical experiments in a Pekeris ocean waveguide are presented that demonstrate that this compressed version of MFP is as effective as traditional MFP even when the compression is significant. The results are particularly promising in the broadband regime where using as few as two random backpropagations per frequency performs almost as well as the traditional broadband MFP but with the added benefit of generic applicability. That is, the computationally intensive backpropagations may be computed offline independently from the received signals, and may be reused to locate any source within the search grid area.

Data driven source localization using a library of nearby shipping sources of opportunity.

A library of broadband (100-1000 Hz) channel impulse responses (CIRs) estimated between a short bottom-mounted vertical line array (VLA) in the Santa Barbara channel and selected locations along the tracks of 27 isolated transiting ships, cumulated over nine days, is constructed using the ray-based blind deconvolution algorithm. Treating this CIR library either as data-derived replica for broadband matched-field processing (MFP) or training data for machine learning yields comparable ranging accuracy (∼50 m) for nearby vessels up to 3.2 km for both methods. Using model-based replica of the direct path only computed for an average sound-speed profile comparatively yields∼110 m ranging accuracy.

STAN: spatio-temporal attention network for pandemic prediction using real-world evidence.

OBJECTIVE: We aim to develop a hybrid model for earlier and more accurate predictions for the number of infected cases in pandemics by (1) using patients' claims data from different counties and states that capture local disease status and medical resource utilization; (2) utilizing demographic similarity and geographical proximity between locations; and (3) integrating pandemic transmission dynamics into a deep learning model. MATERIALS AND METHODS: We proposed a spatio-temporal attention network (STAN) for pandemic prediction. It uses a graph attention network to capture spatio-temporal trends of disease dynamics and to predict the number of cases for a fixed number of days into the future. We also designed a dynamics-based loss term for enhancing long-term predictions. STAN was tested using both real-world patient claims data and COVID-19 statistics over time across US counties. RESULTS: STAN outperforms traditional epidemiological models such as susceptible-infectious-recovered (SIR), susceptible-exposed-infectious-recovered (SEIR), and deep learning models on both long-term and short-term predictions, achieving up to 87% reduction in mean squared error compared to the best baseline prediction model. CONCLUSIONS: By combining information from real-world claims data and disease case counts data, STAN can better predict disease status and medical resource utilization.

STAN: Spatio-Temporal Attention Network for Pandemic Prediction Using Real World Evidence.

OBJECTIVE: The COVID-19 pandemic has created many challenges that need immediate attention. Various epidemiological and deep learning models have been developed to predict the COVID-19 outbreak, but all have limitations that affect the accuracy and robustness of the predictions. Our method aims at addressing these limitations and making earlier and more accurate pandemic outbreak predictions by (1) using patients' EHR data from different counties and states that encode local disease status and medical resource utilization condition; (2) considering demographic similarity and geographical proximity between locations; and (3) integrating pandemic transmission dynamics into deep learning models. MATERIALS AND METHODS: We proposed a spatio-temporal attention network (STAN) for pandemic prediction. It uses an attention-based graph convolutional network to capture geographical and temporal trends and predict the number of cases for a fixed number of days into the future. We also designed a physical law-based loss term for enhancing long-term prediction. STAN was tested using both massive real-world patient data and open source COVID-19 statistics provided by Johns Hopkins university across all U.S. counties. RESULTS: STAN outperforms epidemiological modeling methods such as SIR and SEIR and deep learning models on both long-term and short-term predictions, achieving up to 87% lower mean squared error compared to the best baseline prediction model. CONCLUSIONS: By using information from real-world patient data and geographical data, STAN can better capture the disease status and medical resource utilization information and thus provides more accurate pandemic modeling. With pandemic transmission law based regularization, STAN also achieves good long-term prediction performance.

Sparse Coding Using the Locally Competitive Algorithm on the TrueNorth Neurosynaptic System.

The Locally Competitive Algorithm (LCA) is a biologically plausible computational architecture for sparse coding, where a signal is represented as a linear combination of elements from an over-complete dictionary. In this paper we map the LCA algorithm on the brain-inspired, IBM TrueNorth Neurosynaptic System. We discuss data structures and representation as well as the architecture of functional processing units that perform non-linear threshold, vector-matrix multiplication. We also present the design of the micro-architectural units that facilitate the implementation of dynamical based iterative algorithms. Experimental results with the LCA algorithm using the limited precision, fixed-point arithmetic on TrueNorth compare favorably with results using floating-point computations on a general purpose computer. The scaling of the LCA algorithm within the constraints of the TrueNorth is also discussed.

Extracting the Principal Shape Components via Convex Programming.

We present a general method for extracting a region from an image (or 3D object) that can be expressed, or approximated, by taking unions and set differences from a collection of template shapes in a dictionary. We build on recent work that shows how this geometric problem can be recast in the language of linear algebra, with set operations on shapes translated into linear combinations of vectors, and solved using convex programming. This paper presents a set of sufficient conditions for which this convex program returns the "correct" shape. These conditions are robust in that they can account for the shapes that have indistinct boundaries, or model mismatch between the shapes in the dictionary and the target region in the image. We also present two different methods for solving the convex extraction program. The first method simply recasts the problem as a linear program, while the second uses the alternating direction method of multipliers with a series of easily computed proximal operators. We present a number of numerical experiments that use the framework to perform image segmentation, optical character recognition, and find multi-resolution geometrical descriptions of 3D objects.

Wavelet-domain approximation and compression of piecewise smooth images.

The wavelet transform provides a sparse representation for smooth images, enabling efficient approximation and compression using techniques such as zerotrees. Unfortunately, this sparsity does not extend to piecewise smooth images, where edge discontinuities separating smooth regions persist along smooth contours. This lack of sparsity hampers the efficiency of wavelet-based approximation and compression. On the class of images containing smooth C2 regions separated by edges along smooth C2 contours, for example, the asymptotic rate-distortion (R-D) performance of zerotree-based wavelet coding is limited to D(R) (< or = 1/R, well below the optimal rate of 1/R2. In this paper, we develop a geometric modeling framework for wavelets that addresses this shortcoming. The framework can be interpreted either as 1) an extension to the "zerotree model" for wavelet coefficients that explicitly accounts for edge structure at fine scales, or as 2) a new atomic representation that synthesizes images using a sparse combination of wavelets and wedgeprints--anisotropic atoms that are adapted to edge singularities. Our approach enables a new type of quadtree pruning for piecewise smooth images, using zerotrees in uniformly smooth regions and wedgeprints in regions containing geometry. Using this framework, we develop a prototype image coder that has near-optimal asymptotic R-D performance D(R) < or = (log R)2 /R2 for piecewise smooth C2/C2 images. In addition, we extend the algorithm to compress natural images, exploring the practical problems that arise and attaining promising results in terms of mean-square error and visual quality.

Inferring phage-bacteria infection networks from time-series data.

In communities with bacterial viruses (phage) and bacteria, the phage-bacteria infection network establishes which virus types infect which host types. The structure of the infection network is a key element in understanding community dynamics. Yet, this infection network is often difficult to ascertain. Introduced over 60 years ago, the plaque assay remains the gold standard for establishing who infects whom in a community. This culture-based approach does not scale to environmental samples with increased levels of phage and bacterial diversity, much of which is currently unculturable. Here, we propose an alternative method of inferring phage-bacteria infection networks. This method uses time-series data of fluctuating population densities to estimate the complete interaction network without having to test each phage-bacteria pair individually. We use in silico experiments to analyse the factors affecting the quality of network reconstruction and find robust regimes where accurate reconstructions are possible. In addition, we present a multi-experiment approach where time series from different experiments are combined to improve estimates of the infection network. This approach also mitigates against the possibility of evolutionary changes to relevant phenotypes during the time course of measurement.

Terahertz time-gated spectral imaging for content extraction through layered structures.

Spatial resolution, spectral contrast and occlusion are three major bottlenecks for non-invasive inspection of complex samples with current imaging technologies. We exploit the sub-picosecond time resolution along with spectral resolution provided by terahertz time-domain spectroscopy to computationally extract occluding content from layers whose thicknesses are wavelength comparable. The method uses the statistics of the reflected terahertz electric field at subwavelength gaps to lock into each layer position and then uses a time-gated spectral kurtosis to tune to highest spectral contrast of the content on that specific layer. To demonstrate, occluding textual content was successfully extracted from a packed stack of paper pages down to nine pages without human supervision. The method provides over an order of magnitude enhancement in the signal contrast and can impact inspection of structural defects in wooden objects, plastic components, composites, drugs and especially cultural artefacts with subwavelength or wavelength comparable layers.

Convergence and rate analysis of neural networks for sparse approximation.

We present an analysis of the Locally Competitive Algorithm (LCA), which is a Hopfield-style neural network that efficiently solves sparse approximation problems (e.g., approximating a vector from a dictionary using just a few nonzero coefficients). This class of problems plays a significant role in both theories of neural coding and applications in signal processing. However, the LCA lacks analysis of its convergence properties, and previous results on neural networks for nonsmooth optimization do not apply to the specifics of the LCA architecture. We show that the LCA has desirable convergence properties, such as stability and global convergence to the optimum of the objective function when it is unique. Under some mild conditions, the support of the solution is also proven to be reached in finite time. Furthermore, some restrictions on the problem specifics allow us to characterize the convergence rate of the system by showing that the LCA converges exponentially fast with an analytically bounded convergence rate. We support our analysis with several illustrative simulations.