Gradient Of Objective Function Research Articles

Adjoint method acceleration protocols for model maturation to update static models with time-lapse reservoir surveillance data

We develop an ensemble model-maturation method that is based on the Randomized Maximum Likelihood (RML) technique and adjoint-based computation of objective function gradients. The new approach is especially relevant for rich data sets with time-lapse information content. The inversion method that solves the model-maturation problem takes advantage of the adjoint-based computation of objective function gradients for a very large number of model parameters at the cost of a forward and a backward (adjoint) simulation. The inversion algorithm calibrates model parameters to arbitrary types of production data including time-lapse reservoir-pressure traces by use of a weighted and regularized objective function. We have also developed a new and effective multigrid preconditioning protocol for accelerated iterative linear solutions of the adjoint-simulation step for models with multiple levels of local grid refinement. The protocol is based on a geometric multigrid (GMG) preconditioning technique. Within the model-maturation workflow, a deep-learning technique is applied to establish links between the mesh-based inversion results (e.g., permeability-multiplier fields) and geologic modeling parameters inside a static model (e.g., object dimensions, etc.). Our workflow integrates the learnings from inversion back into the static model, and thereby, ensures the geologic consistency of the static model while improving the quality of the ensuing dynamic model in terms of honoring production and time-lapse data, and reducing forecast uncertainty. This use of deep learning (DL) to post-process the model-maturation outcome effectively converts the conventional continuous-parameter history-matching result into a discrete tomographic inversion result constrained to geological rules encoded in training images.We demonstrate the practical utilization of the adjoint-based model-maturation method on a large time-lapse reservoir-pressure data set using an ensemble of full-field models from a reservoir case study. The model-maturation technique effectively identifies the permeability modification zones that are consistent with alternative geological interpretations and proposes updates to the static model. Upon these updates, the model not only agrees better with the time-lapse reservoir-pressure data but also better honors the tubing-head pressure as well as production logging data. We also provide computational performance indicators that demonstrate the accelerated convergence characteristics of the new iterative linear solver for adjoint equations.

A Single Timescale Stochastic Approximation Method for Nested Stochastic Optimization

We study constrained nested stochastic optimization problems in which the objective function is a composition of two smooth functions whose exact values and derivatives are not available. We propose a single time-scale stochastic approximation algorithm, which we call the Nested Averaged Stochastic Approximation (NASA), to find an approximate stationary point of the problem. The algorithm has two auxiliary averaged sequences (filters) which estimate the gradient of the composite objective function and the inner function value. By using a special Lyapunov function, we show that NASA achieves the sample complexity of ${\cal O}(1/\epsilon^{2})$ for finding an $\epsilon$-approximate stationary point, thus outperforming all extant methods for nested stochastic approximation. Our method and its analysis are the same for both unconstrained and constrained problems, without any need of batch samples for constrained nonconvex stochastic optimization. We also present a simplified variant of the NASA method for solving constrained single level stochastic optimization problems, and we prove the same complexity result for both unconstrained and constrained problems.

Fast extremum seeking using multisine dither and online complex curve fitting

Fast online optimization of uncertain Wiener systems using extremum seeking control (ESC) is investigated. Derivative estimation in extremum seeking is herefore described as an online parametric system identification problem. Multisine dithering is applied with frequencies around the first resonance frequency of the system to remove the time scale separation between dither and plant dynamics which is commonly required in ESC. Recursive use of the Fourier transform, over a moving window of historic data, provides a frequency response function estimate of the system’s local best linear approximation. Continuous online complex curve fitting is then applied to extrapolate to an estimate of the steady-state response which coincides with the local gradient of the steady-state objective function. An analysis of the closed-loop dynamics is provided. Transient improvements and robustness of the approach against plant variation are demonstrated with a simulation example.

Multilayer Topology Optimization of Wideband SIW-to-Waveguide Transitions

This article utilizes a topology optimization approach to design planar multilayer transitions between substrate integrated waveguides (SIWs) and rectangular waveguides (RWGs). The optimization problem is formulated based on the modal field analyses and Maxwell’s equations in the time domain solved by the finite-difference time-domain (FDTD) method. We present a time-domain boundary condition based on the Klein–Gordon equation to split traveling waves at homogeneous waveguide ports. We employ the boundary condition to compute portal quantities and to devise an adjoint-field system that enabled an efficient computation of the objective function gradient. We solve design problems that include more than 105 000 design variables by using less than 400 solutions of Maxwell’s equations. Moreover, a new formulation that effectively combats the development of in-band resonances in the design is presented. The transition configuration allows the direct mount of conventional RWG sections on the circuit board and aims to cover the entire K-band. The guiding structure of the optimized transition requires blind vias, which is realized by a simple and cost-efficient technique. In addition, the transition is optimized for three different setups that can be used to provide different field polarizations. The proposed transitions show less than 1-dB insertion loss and around 15-dB return loss over the frequency interval 18–28 GHz. Several prototypes are fabricated with an excellent match between the simulation and measurement results.

Distributed Continuous-Time and Discrete-Time Optimization With Nonuniform Unbounded Convex Constraint Sets and Nonuniform Stepsizes

This paper is devoted to distributed continuous-time and discrete-time optimization problems with nonuniform convex constraint sets and nonuniform stepsizes for general differentiable convex objective functions. The communication graphs are not required to be strongly connected at any time, the gradients of the local objective functions are not required to be bounded when their independent variables tend to infinity, and the constraint sets are not required to be bounded. For continuous-time multi-agent systems, a distributed continuous algorithm is first introduced where the stepsizes and the convex constraint sets are both nonuniform. It is shown that all agents reach a consensus while minimizing the team objective function even when the constraint sets are unbounded. After that, the obtained results are extended to discrete-time multi-agent systems and then the case where each agent remains in a corresponding convex constraint set is studied. To ensure all agents to remain in a bounded region, a switching mechanism is introduced in the algorithms. It is shown that the distributed optimization problems can be solved, even though the discretization of the algorithms might deviate the convergence of the agents from the minimum of the objective functions. Finally, numerical examples are included to show the obtained theoretical results.

Transmissivity Identification by Combination of CVFEM and Genetic Algorithm: Application to the Coastal Aquifer

The solution of inverse problems in groundwater flow has been massively invested by several researchers around the world. This type of problem has been formulated by a constrained optimization problem and this constraint is none other than the direct problem (DP) itself. Thus, solving algorithms are developed that simultaneously solve the direct problem (Darcy’s equation) and the associated optimization problem. Several papers have been published in the literature using optimization methods based on computation of the objective function gradients. This type of method suffers from the inability to provide a global optimum. Similarly, they also have the disadvantage of not being applicable to objective functions of discontinuous derivatives. This paper is proposed to avoid these disadvantages. Indeed, for the optimization phase, we use random search-based methods that do not use derivative computations, but based on a search step followed only by evaluation of the objective function as many times as necessary to the convergence towards the global optimum. Among the different algorithms of this type of methods, we adopted the genetic algorithm (GA). On the other hand, the numerical solution of the direct problem is accomplished by the CVFEM discretization method (Control Volume Finite Element Method) which ensures the mass conservation in a natural way by its mathematical formulation. The resulting computation code HySubF-CVFEM (Hydrodynamic of Subsurface Flow by Control Volume Finite Element Method) solves the Darcy equation in a heterogeneous porous medium. Thus, this paper describes the description of the integrated optimization algorithm called HySubF-CVFEM/GA that has been successfully implemented and validated successfully compared to a schematic flow case offering analytical solutions. The results of this comparison are qualified of excellent accuracy. To identify the transmissivity field of the realistic study area, the code HySubF-CVFEM/GA was applied to the coastal “Chaouia” groundwater located in Western of Morocco. This aquifer of high heterogeneity is essential for water resources for the Casablanca region. Results analysis of this study has shown that the developed code is capable of providing high accuracy transmissivity fields, thus representing the heterogeneity observed in situ. However, in comparison with gradient method optimization the HySubF-CVFEM/GA code converges too slowly to the optimal solution (large CPU-time consuming). Despite this disadvantage, and given the high accuracy of the obtained results, the HySubF-CVFEM/GA code can be recommended to solve in an efficient and effective manner the identification parameters problems in hydrogeology.

Efficient parameterization of large-scale dynamic models based on relative measurements.

Mechanistic models of biochemical reaction networks facilitate the quantitative understanding of biological processes and the integration of heterogeneous datasets. However, some biological processes require the consideration of comprehensive reaction networks and therefore large-scale models. Parameter estimation for such models poses great challenges, in particular when the data are on a relative scale. Here, we propose a novel hierarchical approach combining (i) the efficient analytic evaluation of optimal scaling, offset and error model parameters with (ii) the scalable evaluation of objective function gradients using adjoint sensitivity analysis. We evaluate the properties of the methods by parameterizing a pan-cancer ordinary differential equation model (>1000 state variables, >4000 parameters) using relative protein, phosphoprotein and viability measurements. The hierarchical formulation improves optimizer performance considerably. Furthermore, we show that this approach allows estimating error model parameters with negligible computational overhead when no experimental estimates are available, providing an unbiased way to weight heterogeneous data. Overall, our hierarchical formulation is applicable to a wide range of models, and allows for the efficient parameterization of large-scale models based on heterogeneous relative measurements. Supplementary code and data are available online at http://doi.org/10.5281/zenodo.3254429 and http://doi.org/10.5281/zenodo.3254441. Supplementary data are available at Bioinformatics online.

Iterative multiscale gradient computation for heterogeneous subsurface flow

We introduce a semi-analytical iterative multiscale derivative computation methodology that allows for error control and reduction to any desired accuracy, up to fine-scale precision. The model responses are computed by the multiscale forward simulation of flow in heterogeneous porous media. The derivative computation method is based on the augmentation of the model equation and state vectors with the smoothing stage defined by the iterative multiscale method. In the formulation, we avoid additional complexity involved in computing partial derivatives associated to the smoothing step. We account for it as an approximate derivative computation stage. The numerical experiments illustrate how the newly introduced derivative method computes misfit objective function gradients that converge to fine-scale one as the iterative multiscale residual converges. The robustness of the methodology is investigated for test cases with high contrast permeability fields. The iterative multiscale gradient method casts a promising approach, with minimal accuracy-efficiency tradeoff, for large-scale heterogeneous porous media optimization problems.

A fast inverse direct aperture optimization algorithm for intensity-modulated radiation therapy.

The goal of this work was to develop and evaluate a fast inverse direct aperture optimization (FIDAO) algorithm for IMRT treatment planning and plan adaptation. A previously proposed fluence map optimization algorithm called fast inverse dose optimization (FIDO) was extended to optimize the aperture shapes and weights of IMRT beams. FIDO is a very fast fluence map optimization algorithm for IMRT that finds the global minimum using direct matrix inversion without unphysical negative beam weights. In this study, an equivalent second-order Taylor series expansion of the FIDO objective function was used, which allowed for the objective function value and gradient vector to be computed very efficiently during direct aperture optimization, resulting in faster optimization. To evaluate the speed gained with FIDAO, a proof-of-concept algorithm was developed in MATLAB using an interior-point optimization method to solve the reformulated aperture-based FIDO problem. The FIDAO algorithm was used to optimize four step-and-shoot IMRT cases: on the AAPM TG-119 phantom as well as a liver, prostate, and head-and-neck clinical cases. Results were compared with a conventional DAO algorithm that uses the same interior-point method but using the standard formulation of the objective function and its gradient vector. A substantial gain in optimization speed was obtained with the prototype FIDAO algorithm compared to the conventional DAO algorithm while producing plans of similar quality. The optimization time (number of iterations) for the prototype FIDAO algorithm vs the conventional DAO algorithm was 0.3s (17) vs 56.7s (50); 2.0s (28) vs 134.1s (57); 2.5s (26) vs 180.6s (107); and 6.7s (20) vs 469.4s (482) in the TG-119 phantom, liver, prostate, and head-and-neck examples, respectively. A new direct aperture optimization algorithm based on FIDO was developed. For the four IMRT test cases examined, this algorithm executed approximately 70-200 times faster without compromising the IMRT plan quality.

Source-free converted-wave reverse time migration: Formulation and limitations

The source-free converted-wave (SFCW) reverse time migration is an alternative way to image the subsurface when the source information is missing or the overburden velocities are complex. However, it is challenging due to the lack of source-wavefield constraints. We have derived the SFCW imaging condition from the first gradients of waveform matching objective functions using a set of coupled P- and S-potential equations and small-perturbation approximations. We found that the images obtained from this SFCW imaging condition are second-order approximations to the shear-modulus perturbations. The images produced by the proposed imaging condition are free of the polarity reversal effects and have clear physical meanings. We also find that the acoustic wave equations can be used to back propagate the recorded P- and S-potentials, and the resulting images have the same kinematic accuracy and fewer unphysical mode-conversion artifacts than the SFCW images obtained by the elastic wave equations. Using multiple numerical examples, we determine the accuracy of our formulations, analyze the resolution and artifacts on the resulting images, and develop the limitations of the proposed method.

Source-independent seismic envelope inversion based on the direct envelope Fréchet derivative

Seismic envelope inversion (EI) uses low-frequency envelope data to recover long-wavelength components of the subsurface media. Conventional EI uses the same waveform Fréchet derivative as conventional full-waveform inversion. Due to linearization of the sensitivity operator (Born approximation), neither of these methods can yield good inversion results for media with strong preturbations, such as salt domes, when the source lacks low-frequency information. Because seismic envelope data contain large amount of ultra-low-frequency information and the direct envelope Fréchet derivative maps envelope data perturbation directly to velocity perturbation, the direct envelope inversion (DEI) method (based on the direct envelope Fréchet derivative) can handle such strong nonlinear inversion problems. However, this method is sensitive to source wavelet errors. We developed a source-independent DEI method. To achieve the source-independent objective function, we derive a convolution expression for the envelope data. We derive the gradient of the new objective function by using the direct envelope Fréchet derivative. Numerical tests conducted on a 2D salt model indicate that our method can achieve good reconstruction of salt bodies (strong velocity perturbations) and recover low-velocity background structures (weak velocity perturbations), despite using an inaccurate source wavelet.

A general B-Spline surfaces theoretical framework for optimisation of variable angle-tow laminates

This work deals with the optimum design of variable angle-tow (VAT) composites in the theoretical framework of B-Spline surfaces. The design problem is formulated as a constrained optimisation problem in the context of the multi-scale two-level (MS2L) optimisation methodology. The proposed approach aims at optimising all parameters characterising the VAT laminate at each pertinent scale (mesoscopic and macroscopic ones). In particular, this study focuses on the first level of the MS2L optimisation strategy that aims at determining, at the VAT laminate macroscopic scale, the optimum distribution of the laminate mechanical properties over the structure. At this stage, the behaviour of the VAT composite is described by means of the polar formalism and the spatial distribution of the polar parameters is represented through a set of suitable B-Spline surfaces. The expression of the gradient of objective and constraint functions is determined in a closed form by exploiting the properties of the B-Spline surfaces. Moreover, the effect of the discrete variables tuning the shape of the B-Spline surfaces (mainly the number of control points) on the quality of the optimum solution is investigated. The effectiveness of the proposed approach is tested through some meaningful benchmarks taken from literature.

An efficient nonlinear Fresnel volume tomography using an improved scattering-integral algorithm

A new nonlinear Fresnel volume tomography (FVT) method is proposed in which a gradient-guided approach is used to replace the conventional equation-solving method. An improved scattering-integral algorithm is used to compute the objective function gradient with respect to the model parameters and the diagonal Hessian. The key principle of this algorithm is the transformation of the gradient computation into an accumulation of multiple vector-scalar products, where each vector is the corresponding individual Fresnel volume kernel of a source-receiver pair. This implementation has several advantages over conventional equation-solving approaches. Firstly, neither the Fresnel volumes nor the Hessian matrix must be stored in memory in advance, thereby ensuring low memory requirements. Secondly, no large equations must be iteratively solved using a general inverse algorithm such as least-squares QR decomposition (LSQR), thereby ensuring a highly efficient inversion procedure. Thirdly, the steepest-descent direction is used to ensure stable convergence. Finally, and most importantly, the preconditioning of the steepest-descent direction, which is equivalent to achieving illumination compensation, can be conveniently performed. This proposed nonlinear FVT method is applied to both synthetic and field data to invert finite-frequency traveltimes of first-arrival waveforms for the near-surface velocity structure. Shallow accuracies comparable to or better than those obtained using the equation-solving FVT or ray-based tomography are achieved. Furthermore, both the memory requirements and computation time are reduced and deeper inversion results are obtained using the diagonal Hessian.

Experimental research on noninvasive reconstruction of optical parameter fields based on transient radiative transfer equation for diagnosis applications

The noninvasive reconstruction of optical parameter fields in participating phantom is investigated experimentally based on the measured time-resolved radiative signals, which are detected by the time-correlated single-photon counting system. The discrete ordinate method is employed to solve the forward transient radiative transfer equation to simulate the time-resolved radiative transfer in the physical phantom exposed to ultra-short pulse laser irradiation. On top of that, the sequential quadratic programming algorithm based on the generalized Gaussian Markov random field is applied to solve the inverse problem. To improve the computational efficiency, an adjoint equation technique is adopted to determine the gradient of objective function. Good agreement was obtained between the predicted optical parameter fields and realistic ones. All the reconstruction results demonstrate that the proposed reconstruction methods are proved to be accurate and efficient experimentally.

DNN-Based Source Enhancement to Increase Objective Sound Quality Assessment Score

We propose a training method for deep neural network (DNN)-based source enhancement to increase objective sound quality assessment (OSQA) scores such as the perceptual evaluation of speech quality (PESQ). In many conventional studies, DNNs have been used as a mapping function to estimate time-frequency masks and trained to minimize an analytically tractable objective function such as the mean squared error (MSE). Since OSQA scores have been used widely for sound-quality evaluation, constructing DNNs to increase OSQA scores would be better than using the minimum-MSE to create high-quality output signals. However, since most OSQA scores are not analytically tractable, \textit{i.e.}, they are black boxes, the gradient of the objective function cannot be calculated by simply applying back-propagation. To calculate the gradient of the OSQA-based objective function, we formulated a DNN optimization scheme on the basis of \textit{black-box optimization}, which is used for training a computer that plays a game. For a black-box-optimization scheme, we adopt the policy gradient method for calculating the gradient on the basis of a sampling algorithm. To simulate output signals using the sampling algorithm, DNNs are used to estimate the probability density function of the output signals that maximize OSQA scores. The OSQA scores are calculated from the simulated output signals, and the DNNs are trained to increase the probability of generating the simulated output signals that achieve high OSQA scores. Through several experiments, we found that OSQA scores significantly increased by applying the proposed method, even though the MSE was not minimized.

Elastic reverse time migration using acoustic propagators

Elastic wave imaging has been a significant challenge in the exploration industry due to the complexities in wave physics and numerical implementation. We have separated the governing equations for P- and S-wave propagation without the assumptions of homogeneous Lamé parameters to capture the mode conversion between the two body waves in an isotropic, constant-density medium. The resulting set of two coupled second-order equations for P- and S-potentials clearly demonstrates that mode conversion only occurs at the discontinuities of the shear modulus. Applying the Born approximation to the new equations, we derive the PP, PS, SP, and SS imaging conditions from the first gradients of waveform matching objective functions. The resulting images are consistent with the physical perturbations of the elastic parameters, and, hence, they are automatically free of the polarity reversal artifacts in the converted images. When implementing elastic reverse time migration (RTM), we find that scalar wave equations can be used to back propagate the recorded P-potential, as well as individual components in the vector field of the S-potential. Compared with conventional elastic RTM, the proposed elastic RTM implementation using acoustic propagators not only simplifies the imaging condition, it but also reduces the computational cost and the artifacts in the images. We have determined the accuracy of our method using 2D and 3D numerical examples.

Adaptive quantum genetic algorithm for task sequence planning of complex assembly systems

Task sequence planning (TSP) is the key factor to the efficiency, stableness, and cost of a complex assembly system. To address the issue, an adaptive quantum genetic algorithm based on artificial potential field and gradient of object function is proposed to optimise the solving process, and to obtain the optimal TSP scheme. The simulation results indicate that the proposed algorithm can perform higher efficiency and stableness than the previously reported methods.

Distributed Adaptive Convex Optimization on Directed Graphs via Continuous-Time Algorithms

This note considers the distributed optimization problem on directed graphs with nonconvex local objective functions and the unknown network connectivity. A new adaptive algorithm is proposed to minimize a differentiable global objective function. By introducing dynamic coupling gains and updating the coupling gains using relative information of system states, the nonconvexity of local objective functions, unknown network connectivity, and the uncertain dynamics caused by locally Lipschitz gradients are tackled concurrently. Consequently, the global asymptotic convergence is established when the global objective function is strongly convex and the gradients of local objective functions are only locally Lipschitz. When the communication graph is strongly connected and weight-balanced, the algorithm is independent of any global information. Then, the algorithm is naturally extended to unbalanced directed graphs by using the left eigenvector of the Laplacian matrix associated with the zero eigenvalue. Several numerical simulations are presented to verify the results.

Data Poisoning Attacks on Multi-Task Relationship Learning

Multi-task learning (MTL) is a machine learning paradigm that improves the performance of each task by exploiting useful information contained in multiple related tasks. However, the relatedness of tasks can be exploited by attackers to launch data poisoning attacks, which has been demonstrated a big threat to single-task learning. In this paper, we provide the first study on the vulnerability of MTL. Specifically, we focus on multi-task relationship learning (MTRL) models, a popular subclass of MTL models where task relationships are quantized and are learned directly from training data. We formulate the problem of computing optimal poisoning attacks on MTRL as a bilevel program that is adaptive to arbitrary choice of target tasks and attacking tasks. We propose an efficient algorithm called PATOM for computing optimal attack strategies. PATOM leverages the optimality conditions of the subproblem of MTRL to compute the implicit gradients of the upper level objective function. Experimental results on real-world datasets show that MTRL models are very sensitive to poisoning attacks and the attacker can significantly degrade the performance of target tasks, by either directly poisoning the target tasks or indirectly poisoning the related tasks exploiting the task relatedness. We also found that the tasks being attacked are always strongly correlated, which provides a clue for defending against such attacks.

Non-destructive testing of ceramic materials using mid-infrared ultrashort-pulse laser

The non-destructive testing (NDT) of ceramic materials using mid-infrared ultrashort-pulse laser is investigated in this study. The discrete ordinate method is applied to solve the transient radiative transfer equation in 2D semitransparent medium and the emerging radiative intensity on boundary serves as input for the inverse analysis. The sequential quadratic programming algorithm is employed as the inverse technique to optimize objective function, in which the gradient of objective function with respect to reconstruction parameters is calculated using the adjoint model. Two reticulated porous ceramics including partially stabilized zirconia and oxide-bonded silicon carbide are tested. The retrieval results show that the main characteristics of defects such as optical properties, geometric shapes and positions can be accurately reconstructed by the present model. The proposed technique is effective and robust in NDT of ceramics even with measurement errors.