Gradient-based Analysis Research Articles

Machine-learning-based water quality management of river with serial impoundments in the Republic of Korea

Euiam Lake in the Republic of Korea This study establishes a framework to prioritize total phosphorus (TP) management strategies based on machine learning (ML). A comparative analysis is conducted to evaluate the performance of four ML methods: random forest (RF), extreme gradient boosting (XGBoost), deep neural network (DNN), and long short-term memory (LSTM). The LSTM-based model is selected as the optimal predictive model of TP concentration in Euiam Lake (E_TP) on seasons (May to October) with high rainfall and inflow from two upstream dams (Chuncheon Dam and Soyanggang Dam). We also perform a gradient-based analysis to figure out the most influential factors on E_TP using the LSTM model. The top four priority factors are TP concentrations and suspended solids concentrations in the upstream dams. This application of the gradient-based analysis enables the predictive model to discuss quantitative reductions in the priorities. Based on these numerical results, we anticipate that the proposed framework can enhance the feasibility of management practices for achieving the water quality management goal of the study region. This study demonstrates that a robust predictive model can be developed for a serial impoundment system with distinct seasonal characteristics of rainfall, temperature, and water quality, thereby facilitating the selection of management priorities. Based on the predictive model results, we conclude that it is the key for managing the target TP concentration to prioritize the incoming TP concentrations and determine the quantitative • Water quality management strategy is established using RF, XGBoost, DNN, and LSTM. • The LSTM model outperforms the other ML methods when predicting the TP concentration in Euiam Lake. • The LSTM gradient-based analysis determines TP management priorities and their quantitative reduction.

Developing and Evaluating Deep Neural Network-Based Denoising for Nanoparticle TEM Images with Ultra-Low Signal-to-Noise

Abstract A deep convolutional neural network has been developed to denoise atomic-resolution transmission electron microscope image datasets of nanoparticles acquired using direct electron counting detectors, for applications where the image signal is severely limited by shot noise. The network was applied to a model system of CeO2-supported Pt nanoparticles. We leverage multislice image simulations to generate a large and flexible dataset for training the network. The proposed network outperforms state-of-the-art denoising methods on both simulated and experimental test data. Factors contributing to the performance are identified, including (a) the geometry of the images used during training and (b) the size of the network's receptive field. Through a gradient-based analysis, we investigate the mechanisms learned by the network to denoise experimental images. This shows that the network exploits both extended and local information in the noisy measurements, for example, by adapting its filtering approach when it encounters atomic-level defects at the nanoparticle surface. Extensive analysis has been done to characterize the network's ability to correctly predict the exact atomic structure at the nanoparticle surface. Finally, we develop an approach based on the log-likelihood ratio test that provides a quantitative measure of the agreement between the noisy observation and the atomic-level structure in the network-denoised image.

An Efficient and Accurate Iris Recognition Algorithm Based on a Novel Condensed 2-ch Deep Convolutional Neural Network.

Recently, deep learning approaches, especially convolutional neural networks (CNNs), have attracted extensive attention in iris recognition. Though CNN-based approaches realize automatic feature extraction and achieve outstanding performance, they usually require more training samples and higher computational complexity than the classic methods. This work focuses on training a novel condensed 2-channel (2-ch) CNN with few training samples for efficient and accurate iris identification and verification. A multi-branch CNN with three well-designed online augmentation schemes and radial attention layers is first proposed as a high-performance basic iris classifier. Then, both branch pruning and channel pruning are achieved by analyzing the weight distribution of the model. Finally, fast finetuning is optionally applied, which can significantly improve the performance of the pruned CNN while alleviating the computational burden. In addition, we further investigate the encoding ability of 2-ch CNN and propose an efficient iris recognition scheme suitable for large database application scenarios. Moreover, the gradient-based analysis results indicate that the proposed algorithm is robust to various image contaminations. We comprehensively evaluated our algorithm on three publicly available iris databases for which the results proved satisfactory for real-time iris recognition.

Gradient-Based Intraprediction Fusion for Video Coding

Video coding is more important than ever due to emergence of larger resolutions, new formats requiring higher bandwidths, prevalence of new video services, etc. Consequently, new codecs are designed to minimize transmission rates of such contents without sacrificing quality. This article proposes an intraprediction solution for future video codecs that save the bit rates of intramode signaling for luminance and chrominance blocks. In doing so, modes are implicitly derived at the decoder side through a gradient-based analysis. In contrast to state-of-the-art intraprediction that uses a single mode for a luma block, the proposed method offers a combination of multiple derived modes. Furthermore, the chrominance components Cb/Cr are allowed to have their separate modes, instead of sharing one mode. The implementation of the proposed method in the Versatile Video Coding reference software (VTM) provides 0.72% Bjontegaard-delta rate saving, with run-time complexity of 110% and 104% at the encoder and decoder sides, respectively.

Gradient-based model calibration with proxy-model assistance

Use of a proxy model in gradient-based calibration and uncertainty analysis of a complex groundwater model with large run times and problematic numerical behaviour is described. The methodology is general, and can be used with models of all types. The proxy model is based on a series of analytical functions that link all model outputs used in the calibration process to all parameters requiring estimation. In enforcing history-matching constraints during the calibration and post-calibration uncertainty analysis processes, the proxy model is run for the purposes of populating the Jacobian matrix, while the original model is run when testing parameter upgrades; the latter process is readily parallelized. Use of a proxy model in this fashion dramatically reduces the computational burden of complex model calibration and uncertainty analysis. At the same time, the effect of model numerical misbehaviour on calculation of local gradients is mitigated, this allowing access to the benefits of gradient-based analysis where lack of integrity in finite-difference derivatives calculation would otherwise have impeded such access. Construction of a proxy model, and its subsequent use in calibration of a complex model, and in analysing the uncertainties of predictions made by that model, is implemented in the PEST suite.

Object-shape recognition and 3D reconstruction from tactile sensor images

This article presents a novel approach of edged and edgeless object-shape recognition and 3D reconstruction from gradient-based analysis of tactile images. We recognize an object's shape by visualizing a surface topology in our mind while grasping the object in our palm and also taking help from our past experience of exploring similar kind of objects. The proposed hybrid recognition strategy works in similar way in two stages. In the first stage, conventional object-shape recognition using linear support vector machine classifier is performed where regional descriptors features have been extracted from the tactile image. A 3D shape reconstruction is also performed depending upon the edged or edgeless objects classified from the tactile images. In the second stage, the hybrid recognition scheme utilizes the feature set comprising both the previously obtained regional descriptors features and some gradient-related information from the reconstructed object-shape image for the final recognition in corresponding four classes of objects viz. planar, one-edged object, two-edged object and cylindrical objects. The hybrid strategy achieves 97.62% classification accuracy, while the conventional recognition scheme reaches only to 92.60%. Moreover, the proposed algorithm has been proved to be less noise prone and more statistically robust.

Derivatives of Pearson Correlation for Gradient-based Analysis of Biomedical Data

In biomedical analytics one of the major criteria for the characterization of similarities between measured data items is correlation. We demonstrate the use of the formal derivative of Pearson correlation for gradientbased optimization of data models. Firstly, individual data attributes can be rated according to their impact on pairwise data relationships, analogous to the variance measure in Euclidean space. Secondly, a versatile method is presented to perform faithful multi-dimensional scaling from a high-dimensional space of source data to a low-dimensional target space, a method driven by maximization of correlation between distances of static source data and adaptive target vectors. As shown for mass spectroscopy data, a combination of attribute rating and data visualization helps revealing interesting data properties.

Gradient-based analysis of non-Fourier motion

A gradient-based image analysis technique is applied to a class of non-Fourier stimuli. To create the stimuli, n translating sine waves with identical spatial and temporal frequencies, but separated by 2 π/ n radians, are spatially randomly sampled to produce a P n stimulus. For n⩾2, the stimuli are non-Fourier. Local image gradients are represented in the form of a gradient plot, a histogram which shows the frequency of ranges of temporal gradient/spatial gradient pairs occurring. It is shown that the gradient plots contain features, oriented in gradient space, which indicate correct non-Fourier velocity. As n increases, so too does the complexity of the gradient plots, a finding which may account for the concomitant decrease in perceived coherent motion [Vision Res 37 (1997) 1459]. This paper demonstrates that the gradient plot and associated velocity plots are a useful way of assessing gradient-based motion information. Compared to the traditional Fourier based approach, gradient-based analysis can lead to different judgement of the motion information available to standard models of low-level motion processing.

Robust velocity computation from a biologically motivated model of motion perception

Current computational models of motion processing in the primate motion pathway do not cope well with image sequences in which a moving pattern is superimposed upon a static texture. The use of non–linear operations and the need for contrast normalization in motion models mean that the separation of the influences of moving and static patterns on the motion computation is not trivial. Therefore, the response to the superposition of static and moving patterns provides an important means of testing various computational strategies. Here we describe a computational model of motion processing in the visual cortex, one of the advantages of which is that it is highly resistant to interference from static patterns.