Non-local Search Research Articles

Learning real-world heterogeneous noise models with a benchmark dataset

Noise modeling is an important research field in computer vision; the design of an accurate model for imaging sensor noise depends on not only a comprehensive benchmark dataset of the real world, but also a precise design of the noise modeling algorithm. However, due to the inaccurate estimation method of noise-free images and limited shooting scenes, the current realistic datasets could not describe the diverse noise properties sufficiently. Moreover, popular parametric noise models are not sophisticated enough to characterize the real-world noise exactly. In this work, we first construct a more comprehensive dataset of the real world by capturing more indoor and outdoor scenes under different lighting conditions using diverse smartphones, then we propose a non-parametric noise estimation method capable of modeling the spatial heterogeneity of real-world noise patterns. Specifically, in order to characterize the spatial heterogeneity of real-world noise, we assume a non-i.i.d Gaussian distribution and propose a deep convolutional neural network (DCNN)-based approach for learning pixel-wise noise variance maps. To learn the pixel-wise variance map, we have constructed a variance estimation network mapping from the conditional signals (clean image, ISO, and camera model) to surrogate labels obtained from the nonlocal search of similar patches from the clean-noisy image pair. Finally, we conducted denoising and classification experiments using different kinds of simulated noisy images, compared to the Poisson-Gaussian and Noise Flow noise models, the proposed method achieves denoising performance improvements (PSNR) of 1.13 dB and 2.51 dB respectively on the proposed real-world test dataset, denoising and classification results on the real noisy data captured by mobile phones have verified that our approach is more accurate than current noise modeling methods.

Macroscopic limit of a kinetic model describing the switch in T cell migration modes via binary interactions

Experimental results on the immune response to cancer indicate that activation of cytotoxic T lymphocytes (CTLs) through interactions with dendritic cells (DCs) can trigger a change in CTL migration patterns. In particular, while CTLs in the pre-activation state move in a non-local search pattern, the search pattern of activated CTLs is more localised. In this paper, we develop a kinetic model for such a switch in CTL migration modes. The model is formulated as a coupled system of balance equations for the one-particle distribution functions of CTLs in the pre-activation state, activated CTLs and DCs. CTL activation is modelled via binary interactions between CTLs in the pre-activation state and DCs. Moreover, cell motion is represented as a velocity-jump process, with the running time of CTLs in the pre-activation state following a long-tailed distribution, which is consistent with a Lévy walk, and the running time of activated CTLs following a Poisson distribution, which corresponds to Brownian motion. We formally show that the macroscopic limit of the model comprises a coupled system of balance equations for the cell densities, whereby activated CTL movement is described via a classical diffusion term, whilst a fractional diffusion term describes the movement of CTLs in the pre-activation state. The modelling approach presented here and its possible generalisations are expected to find applications in the study of the immune response to cancer and in other biological contexts in which switch from non-local to localised migration patterns occurs.

Accelerating reinforcement learning with a Directional-Gaussian-Smoothing evolution strategy

<p style='text-indent:20px;'>The objective of reinforcement learning (RL) is to find an optimal strategy for solving a dynamical control problem. Evolution strategy (ES) has been shown great promise in many challenging reinforcement learning (RL) tasks, where the underlying dynamical system is only accessible as a black box such that adjoint methods cannot be used. However, existing ES methods have two limitations that hinder its applicability in RL. First, most existing methods rely on Monte Carlo based gradient estimators to generate search directions. Due to low accuracy of Monte Carlo estimators, the RL training suffers from slow convergence and requires more iterations to reach the optimal solution. Second, the landscape of the reward function can be deceptive and may contain many local maxima, causing ES algorithms to prematurely converge and be unable to explore other parts of the parameter space with potentially greater rewards. In this work, we employ a Directional Gaussian Smoothing Evolutionary Strategy (DGS-ES) to accelerate RL training, which is well-suited to address these two challenges with its ability to (i) provide gradient estimates with high accuracy, and (ii) find nonlocal search direction which lays stress on large-scale variation of the reward function and disregards local fluctuation. Through several benchmark RL tasks demonstrated herein, we show that the DGS-ES method is highly scalable, possesses superior wall-clock time, and achieves competitive reward scores to other popular policy gradient and ES approaches.</p>

A computational experiment based on a nonlocal search algorithm using potential theory

We consider a general method to derive an optimization algorithms based on potential theory. This theoretical approach is based on the randomization of an objective function and the computation of directional derivatives of the randomized functional. Using the gradient of the potential field we consider algorithm in terms of the means of the underlying movements in the space of random vectors. Effectively, this process combines the exploration power of random-search algorithms with the exploitation power of direct-search algorithms. At the end we offer the results of a computational expert.

FAMoS: A Flexible and dynamic Algorithm for Model Selection to analyse complex systems dynamics.

Most biological systems are difficult to analyse due to a multitude of interacting components and the concomitant lack of information about the essential dynamics. Finding appropriate models that provide a systematic description of such biological systems and that help to identify their relevant factors and processes can be challenging given the sheer number of possibilities. Model selection algorithms that evaluate the performance of a multitude of different models against experimental data provide a useful tool to identify appropriate model structures. However, many algorithms addressing the analysis of complex dynamical systems, as they are often used in biology, compare a preselected number of models or rely on exhaustive searches of the total model space which might be unfeasible dependent on the number of possibilities. Therefore, we developed an algorithm that is able to perform model selection on complex systems and searches large model spaces in a dynamical way. Our algorithm includes local and newly developed non-local search methods that can prevent the algorithm from ending up in local minima of the model space by accounting for structurally similar processes. We tested and validated the algorithm based on simulated data and showed its flexibility for handling different model structures. We also used the algorithm to analyse experimental data on the cell proliferation dynamics of CD4+ and CD8+ T cells that were cultured under different conditions. Our analyses indicated dynamical changes within the proliferation potential of cells that was reduced within tissue-like 3D ex vivo cultures compared to suspension. Due to the flexibility in handling various model structures, the algorithm is applicable to a large variety of different biological problems and represents a useful tool for the data-oriented evaluation of complex model spaces.

Problemistic search distance and entrepreneurial performance

AbstractResearch SummaryThis paper seeks to extend the problemistic search literature by investigating how far entrepreneurial organizations seeking to improve their chances for success should search, depending on their level of past performance. Drawing on a novel data set from the Google Play app store, the paper finds support for its hypothesis that past performance moderates the relationship between search distance and subsequent performance. As past performance increases, the less beneficial (and potentially more harmful) nonlocal search becomes. While the majority of app developers choose a nonlocal search strategy in response to low first app performance, this is rarely the best choice. Instead, the highest second app performance outcomes are associated with moderate search distances that fall between local and nonlocal extremes.Managerial SummaryThis paper uses a large data set obtained from the Google Play app store to explore the following question: when a nascent app developer's first app performs poorly, how different should its second app be? By comparing the text descriptions of developers' first and second apps, the paper is able to show that as the performance of a first app increases, the more harmful it becomes to make a very different second app. Only at extremely low levels of first app performance is it beneficial for developers to make second apps that are very different from their first apps. In all other cases, making second apps that are moderately different—rather than very similar or very different—is associated with the highest second app performance.

Commercial or Technological Impact? Contextual Determinants of the Value of Nonlocal Search

When does non-local search lead to inventions with high impact? Using US patent data from 1975 to 2017, we address this question by examining how the effects of nonlocal search on technological imp...

Distance to Technology Aspiration and Entry into the Merger Wave

Drawing from the concepts of problemistic search, nonlocal search and merger waves, we investigate the effect of unmet technology aspirations on strategic decisions such as firm acquisitions. Our e...

The search environment is not (always) benign: reassessing the risks of organizational search

AbstractMarch and Simon’s (1958: 50) assumption that search happens within a “benign environment” has become taken-for-granted in organizational studies. The implications of making this assumption are not widely theorized, investigated, or even discussed, and yet, it often appears to have been unwittingly imported into explanations of organizational learning. March and Simon acknowledged that this assumption was not realistic—our work builds on theirs by investigating the origins and consequences of a non-benign search environment. We draw on in-depth qualitative research at a Fortune 500 mining firm to show how perceptions of social risk problematize the assumption of a benign search environment. Causal links are drawn between role equivalence, performance comparisons, and rivalry for social status to explain how social risk is generated and why it can lead actors to view the intraorganizational search environment as non-benign. These perceptions help create what we describe as a “paradox of local equivalence” that leads actors to search for nonlocal solutions. Our causal logic provides a new way of understanding the phenomena of nonlocal search; complements explanations of nonlocal search founded on myopia in organizational learning; and shows how the micro-foundations of existing search models can be adapted to better explain organizational learning. In doing so, this study contributes to recent efforts to improve behavioral explanations of search and learning by bringing the notion of intraorganizational conflict back to center stage in this important area of organizational theory.

Searching in the Non-market Environment: The Impact of Regulatory Search on Firm Innovativeness (WITHDRAWN)

This paper investigated the relationship between regulatory search and firm innovativeness in a contingency framework. First, regulatory search was conceptualized as a non-local and exploratory knowledge acquisition capability in a firm’s non-market environment, specifically, the regulatory environment. Second, a self-developed scale was used to measure regulatory search and the results suggested that the non-local search conducted in the regulatory environment formed two independent factors, namely, reactive and proactive regulatory search, aligning with firms’ non-market strategies of positive anticipation/compliance and proactive participation/influence respectively. Third, a contingency analysis of the regulatory search-innovativeness link conditioned on a firm’s internal resource munificence (proxied by slack) was conducted. We found that slack moderated the relationship between the two regulatory search factors and innovativeness in different manners. Specifically, under a high (low) slack environme...

A non-invasive implementation of a mixed domain decomposition method for frictional contact problems

A non-invasive implementation of the Latin domain decomposition method for frictional contact problems is described. The formulation implies to deal with mixed (Robin) conditions on the faces of the subdomains, which is not a classical feature of commercial software. Therefore we propose a new implementation of the linear stage of the Latin method with a non-local search direction built as the stiffness of a layer of elements on the interfaces. This choice enables us to implement the method within the open source software Code_Aster, and to derive 2D and 3D examples with similar performance as the standard Latin method.

Polarimetric SAR Despeckling by Integrating Stochastic Sampling and Contextual Patch Dissimilarity Exploration

Speckle reduction has been the longstanding task since the invention of synthetic aperture radar (SAR). For further image analysis and interpretation, it usually demands better speckle suppression to preserve spatial and polarimetric information of the polarimetric SAR (PolSAR) images. In this paper, a new PolSAR filtering algorithm which combines stochastic sampling based on nonlocal mean, random walk model, and contextual patch dissimilarity is proposed. The nonlocal mean suppresses the speckle effectively while preserving the details and polarimetric information well. However, the size of nonlocal search window is fixed without taking the homogeneous and heterogeneous areas into consideration. Thus, it usually leads to computational redundancy. Correspondingly, we propose to apply the random walk to reduce the search domain. Contextual patch is also proposed to represent the large surroundings of a patch in a compact fashion. More precisely, the random walk model is utilized to determine the sampling path. Then, the traditional center patch dissimilarity and contextual patch dissimilarity are employed to measure the transition probability of the random walk sequence. Finally, the denoised estimation of PolSAR image is obtained by weight function derived from the transition probability. Since the stochastic sampling combines spatial random walk and the polarimetric dissimilarity measurements, the proposed algorithm fully exploits both the spatial and polarimetric information. Experimental results on synthetic and real PolSAR data demonstrate the effectiveness of the proposed method in terms of spatial and polarimetric information maintenance.

Random-Valued Impulse Noise Removal Using Non-Local Search for Similar Structures and Sparse Representation

In this paper, we introduce a new method to remove random-valued impulse noise in an image. Random-valued impulse noise replaces the pixel value at a random position by a random value. Due to the randomness of the noisy pixel values, it is difficult to detect them by comparison with neighboring pixels, which is used in many conventional methods. Then we improve the recent noise detector which uses a non-local search of similar structure. Next we propose a new noise removal algorithm by sparse representation using DCT basis. Furthermore, the sparse representation can remove impulse noise by using the neighboring similar image patch. This method has much more superior noise removal performance than conventional methods.

Knowledge exploration and innovation: A review and an inverse S-curve proposition

AbstractFirms today thrive on innovation. Knowledge exploration, the nonlocal search for new knowledge beyond the firm’s current expertise, is posited to be critical for innovation. This paper seeks to contribute to the research on knowledge exploration in two ways. First, this paper provides a comprehensive review of key empirical studies on knowledge exploration and innovation. Second, this paper proposes a recombinatory search framework of innovation to reconceptualise extant understanding of knowledge exploration on innovation. This new framework focusses on the evolution of the benefits and costs of knowledge exploration, and puts forward an inverse S-curve proposition between knowledge exploration and innovation. Two company cases, IBM and Procter & Gamble, are then used to illustrate the new proposition.

A stochastically evolving non-local search and solutions to inverse problems with sparse data

Building on a martingale approach to global optimization, a powerful stochastic search scheme for the global optimum of cost functions is proposed using change of measures on the states that evolve as diffusion processes and splitting of the state-space along the lines of a Bayesian game. To begin with, the efficacy of the optimizer, when contrasted with one of the most efficient existing schemes, is assessed against a family of Np-hard benchmark problems. Then, using both simulated and experimental data, potentialities of the new proposal are further explored in the context of an inverse problem of significance in photoacoustic imaging, wherein the superior reconstruction features of a global search vis-à-vis the commonly adopted local or quasi-local schemes are brought into relief.

Toward an aspiration-level theory of open innovation

Although open innovation has become increasingly established in the management literature, comprehensive theoretical explanations of what drives firms to be open are sparse. Taking the perspective of the behavioral theory of the firm, we conceive of open innovation as a form of nonlocal search, arguing that firms will turn to open innovation when substantially under- or over-performing relative to their aspirations. We further enquire how this relationship is moderated by firm-specific innovation-related resources: human capital, research and development investment, and patents. Employing a representative survey of UK firms, we find some evidence of moderation, allowing us to present explanations of search through open innovation and contribute to the behavioral theory of the firm itself.

Secure Nonlocal Denoising in Outsourced Images

Signal processing in the encrypted domain becomes a desired technique to protect privacy of outsourced data in cloud. In this article, we propose a double-cipher scheme to implement nonlocal means (NLM) denoising in encrypted images. In this scheme, one ciphertext is generated by the Paillier scheme, which enables the mean filter, and the other is obtained by a privacy-preserving transform, which enables the nonlocal search. By the privacy-preserving transform, the cloud server can search the similar pixel blocks in the ciphertexts with the same speed as in the plaintexts; thus, the proposed method can be executed fast. To enhance the security, we randomly permutate both ciphertexts. To reduce the denoising complexity caused by random permutation, a random NLM method is exploited in the encrypted domain. The experimental results show that the quality of denoised images in the encrypted domain is comparable to that obtained in the plain domain.

Further advances on Bayesian Ying-Yang harmony learning

After a short tutorial on the fundamentals of Bayes approaches and Bayesian Ying-Yang (BYY) harmony learning, this paper introduces new progresses. A generic information harmonising dynamics of BYY harmony learning is proposed with the help of a Lagrange variety preservation principle, which provides Lagrange-like implementations of Ying-Yang alternative nonlocal search for various learning tasks and unifies attention, detection, problem-solving, adaptation, learning and model selection from an information harmonising perspective. In this framework, new algorithms are developed to implement Ying-Yang alternative nonlocal search for learning Gaussian mixture and several typical exemplars of linear matrix system, including factor analysis (FA), mixture of local FA, binary FA, nonGaussian FA, de-noised Gaussian mixture, sparse multivariate regression, temporal FA and temporal binary FA, as well as a generalised bilinear matrix system that covers not only these linear models but also manifold learning, gene regulatory networks and the generalised linear mixed model. These algorithms are featured with a favourable nature of automatic model selection and a unified formulation in performing unsupervised learning and semi-supervised learning. Also, we propose a principle of preserving multiple convex combinations, which leads alternative search algorithms. Finally, we provide a chronological outline of the history of BYY learning studies.

Toward an aspiration-level theory of open innovation

Although open innovation has become increasingly established in the management literature, comprehensive theoretical explanations of what drives firms to be open are sparse. By taking the perspective of the behavioral theory of the firm, we conceive of open innovation as a form of non-local search, arguing that that firms will use it when substantially under- or overperforming their aspirations. We further enquire how this relationship is moderated by firm-specific innovation-related assets: human capital, R&D investment, and patents. Employ a representative survey of UK firms, we find strong evidence of moderation, allowing us to present explanations of search through open innovation and contribute to the behavioral theory of the firm itself.

Algorithm of multiobjective operational control of microgrids

One of important energy development directions, as evidenced by the global practice, is the intellectualization of generation, transmission and distribution of energy, formation of the so-called microgrids. In this regard, the results of the research, related to developing the method for prompt operational control of the sources of distributed generation of autonomous microgrid are given in the paper. Solving such problems inUkraineis complicated by limited availability of material resources for the fundamental technical modernization of the industry, the lack of adequate information environment, the imperfection of legal and regulatory bases. These circumstances make the issues of forming adequate mathematical models, used to solve the problems of determining the optimal structure of microgrids and their operational control especially relevant. Given this, the necessity of comprehensive consideration of the uncertainty factor, including considering the uncertainty of source information, objectives and conditions, when analyzing the specified problem is substantiated in the paper. For this purpose, a series of objective functions, reflecting the economic, technical and environmental aspects of the microgrid operation, which were originally presented in a linguistic form that allows to take into account both the information uncertainty, and multicriterion nature of the problem is formed. Capacity allocation among individual sources is implemented based on a modified non-local search algorithm.Thus, accounting multicriteriality is performed using the Bellman-Zadeh approach, which allows to obtain a solution, belonging to the region of compromise, where the optimality principle lies in the maximum satisfaction of all purposes. The proposed algorithm allows to flexibly and effectively consider both quantitative and qualitative characteristics, represented, in particular, by linguistic assessments, differentiate the importance degree of individual criteria, which allow to ensure the maximum adequacy and validity of the obtained solutions and, as a consequence, guarantee the actual efficiency of using generating equipment, installed in the microgrid