Loss Constraints Research Articles

Medical image translation using an edge-guided generative adversarial network with global-to-local feature fusion.

In this paper, we propose a framework based deep learning for medical image translation using paired and unpaired training data. Initially, a deep neural network with an encoder-decoder structure is proposed for image-to-image translation using paired training data. A multi-scale context aggregation approach is then used to extract various features from different levels of encoding, which are used during the corresponding network decoding stage. At this point, we further propose an edge-guided generative adversarial network for image-to-image translation based on unpaired training data. An edge constraint loss function is used to improve network performance in tissue boundaries. To analyze framework performance, we conducted five different medical image translation tasks. The assessment demonstrates that the proposed deep learning framework brings significant improvement beyond state-of-the-arts.

Semi-Supervised SAR Target Detection Based on an Improved Faster R-CNN

In the remote sensing image processing field, the synthetic aperture radar (SAR) target-detection methods based on convolutional neural networks (CNNs) have gained remarkable performance relying on large-scale labeled data. However, it is hard to obtain many labeled SAR images. Semi-supervised learning is an effective way to address the issue of limited labels on SAR images because it uses unlabeled data. In this paper, we propose an improved faster regions with CNN features (R-CNN) method, with a decoding module and a domain-adaptation module called FDDA, for semi-supervised SAR target detection. In FDDA, the decoding module is adopted to reconstruct all the labeled and unlabeled samples. In this way, a large number of unlabeled SAR images can be utilized to help structure the latent space and learn the representative features of the SAR images, devoting attention to performance promotion. Moreover, the domain-adaptation module is further introduced to utilize the unlabeled SAR images to promote the discriminability of features with the assistance of the abundantly labeled optical remote sensing (ORS) images. Specifically, the transferable features between the ORS images and SAR images are learned to reduce the domain discrepancy via the mean embedding matching, and the knowledge of ORS images is transferred to the SAR images for target detection. Ultimately, the joint optimization of the detection loss, reconstruction, and domain adaptation constraints leads to the promising performance of the FDDA. The experimental results on the measured SAR image datasets and the ORS images dataset indicate that our method achieves superior SAR target detection performance with limited labeled SAR images.

HPGCN: Hierarchical poselet-guided graph convolutional network for 3D pose estimation

3D pose estimation remains a challenging task since human poses exhibit high ambiguity and multi-granularity. Traditional graph convolution networks (GCNs) accomplish the task by modeling all skeletons as an entire graph, and are unable to fuse combinable part-based features. By observing that human movements occur due to part of human body (i.e. related skeletons and body components, known as the poselet) and those poselets contribute to each movement in a hierarchical fashion, we propose a hierarchical poselet-guided graph convolutional network (HPGCN) for 3D pose estimation from 2D poses. HPGCN sets five primitives of human body as basic poselets, and constitutes high-level poselets according to the kinematic configuration of human body. Moreover, HPGCN forms a fundamental unit by using a diagonally dominant graph convolution layer and a non-local layer, which corporately capture the multi-granular feature of human poses from local to global perspective. Finally HPGCN designs a geometric constraint loss function with constraints on lengths and directions of bone vectors, which help produce reasonable pose regression. We verify the effectiveness of HPGCN on three public 3D human pose benchmarks. Experimental results show that HPGCN outperforms several state-of-the-art methods.

Anatomically constrained squeeze-and-excitation graph attention network for cortical surface parcellation

In order to understand the organizational structures of healthy cerebral cortex and the abnormalities in neurological and psychiatric diseases, it is significant to parcellate the cortical surface. The cortical surface, however, is a highly folded complex geometric structure which challenges automatic cortical surface parcellation. Nowadays, the parcellation methods of cerebral cortex are mostly based on geometric simplification, i.e., iteratively inflating and mapping the cortical surface to a spherical surface for processing, which is time-consuming and cannot make full use of the intrinsic structural information of the original cortical surface. In this study, we proposed an anatomically constrained squeeze-and-excitation graph attention network (ASEGAT) for an end-to-end brain cortical surface parcellation on the original cortical surface manifold. The ASEGAT is formed by two graph attention modules and a squeeze-and-excitation module that incorporate self-attention and head attention for rendering features of each node. Furthermore, we designed an anatomic constraint loss to introduce the anatomical priori of regional adjacency relationships, which could improve the consistency of region labeling. We evaluated our model on a public dataset of 100 manually labeled brain surfaces. Compared with several advanced methods, the results showed that our proposed approach achieved state-of-the-art performance, obtaining an accuracy of 90.65% and a dice score of 89.00%.

Anomaly Detection for Medical Images Using Self-Supervised and Translation-Consistent Features.

As the labeled anomalous medical images are usually difficult to acquire, especially for rare diseases, the deep learning based methods, which heavily rely on the large amount of labeled data, cannot yield a satisfactory performance. Compared to the anomalous data, the normal images without the need of lesion annotation are much easier to collect. In this paper, we propose an anomaly detection framework, namely [Formula: see text], extracting [Formula: see text]elf-supervised and tr [Formula: see text]ns [Formula: see text]ation-consistent features for [Formula: see text]nomaly [Formula: see text]etection. The proposed SALAD is a reconstruction-based method, which learns the manifold of normal data through an encode-and-reconstruct translation between image and latent spaces. In particular, two constraints (i.e., structure similarity loss and center constraint loss) are proposed to regulate the cross-space (i.e., image and feature) translation, which enforce the model to learn translation-consistent and representative features from the normal data. Furthermore, a self-supervised learning module is engaged into our framework to further boost the anomaly detection accuracy by deeply exploiting useful information from the raw normal data. An anomaly score, as a measure to separate the anomalous data from the healthy ones, is constructed based on the learned self-supervised-and-translation-consistent features. Extensive experiments are conducted on optical coherence tomography (OCT) and chest X-ray datasets. The experimental results demonstrate the effectiveness of our approach.

Performance limits of network delay systems based on bandwidth and packet loss constraints

AbstractIn this work, we study the error limits of the network delay systems in terms of bandwidth and packet loss. The tracking error index of networked delay systems is identified by the tracking error energy. The error limits of the network delay systems are achieved by the spectral decomposition and partial decomposition techniques. The results of this work show that the tracking error limits are relevant to the inherent characteristics of the provided object (non‐minimum phase zero, time delay, and unstable pole). In addition, the packet loss, channel noise, and bandwidth also determine the error limits of the network delay systems. The simulation results show that the theory presented is valid.

Table to text generation with accurate content copying

Generating fluent, coherent, and informative text from structured data is called table-to-text generation. Copying words from the table is a common method to solve the “out-of-vocabulary” problem, but it’s difficult to achieve accurate copying. In order to overcome this problem, we invent an auto-regressive framework based on the transformer that combines a copying mechanism and language modeling to generate target texts. Firstly, to make the model better learn the semantic relevance between table and text, we apply a word transformation method, which incorporates the field and position information into the target text to acquire the position of where to copy. Then we propose two auxiliary learning objectives, namely table-text constraint loss and copy loss. Table-text constraint loss is used to effectively model table inputs, whereas copy loss is exploited to precisely copy word fragments from a table. Furthermore, we improve the text search strategy to reduce the probability of generating incoherent and repetitive sentences. The model is verified by experiments on two datasets and better results are obtained than the baseline model. On WIKIBIO, the result is improved from 45.47 to 46.87 on BLEU and from 41.54 to 42.28 on ROUGE. On ROTOWIRE, the result is increased by 4.29% on CO metric, and 1.93 points higher on BLEU.

Cross-Modal Hybrid Feature Fusion for Image-Sentence Matching

Image-sentence matching is a challenging task in the field of language and vision, which aims at measuring the similarities between images and sentence descriptions. Most existing methods independently map the global features of images and sentences into a common space to calculate the image-sentence similarity. However, the image-sentence similarity obtained by these methods may be coarse as (1) an intermediate common space is introduced to implicitly match the heterogeneous features of images and sentences in a global level, and (2) only the inter-modality relations of images and sentences are captured while the intra-modality relations are ignored. To overcome the limitations, we propose a novel Cross-Modal Hybrid Feature Fusion (CMHF) framework for directly learning the image-sentence similarity by fusing multimodal features with inter- and intra-modality relations incorporated. It can robustly capture the high-level interactions between visual regions in images and words in sentences, where flexible attention mechanisms are utilized to generate effective attention flows within and across the modalities of images and sentences. A structured objective with ranking loss constraint is formed in CMHF to learn the image-sentence similarity based on the fused fine-grained features of different modalities bypassing the usage of intermediate common space. Extensive experiments and comprehensive analysis performed on two widely used datasets—Microsoft COCO and Flickr30K—show the effectiveness of the hybrid feature fusion framework in CMHF, in which the state-of-the-art matching performance is achieved by our proposed CMHF method.

Constrained Online Convex Optimization With Feedback Delays

In this article, we study constrained online convex optimization (OCO) in the presence of feedback delays, where a decision maker chooses sequential actions without knowing the loss functions and constraint functions <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a priori</i> . The loss/constraint functions vary with time and their feedback information is revealed to the decision maker with delays, which arise in many applications. We first consider the scenario of delayed function feedback, in which the complete information of the loss/constraint functions is revealed to the decision maker with delays. We develop a modified online saddle point algorithm suitable for constrained OCO with feedback delays. Sublinear regret and sublinear constraint violation bounds are established for the algorithm in terms of the delays. In practice, the complete information (functional forms) of the loss/constraint functions may not be revealed to the decision maker. Thus, we further examine the scenario of delayed bandit feedback, where only the values of the loss/constraint functions at two random points close to the chosen action are revealed to the decision maker with delays. A delayed version of the bandit online saddle point algorithm is proposed by utilizing stochastic gradient estimates of the loss/constraint functions based on delayed bandit feedback. We also establish sublinear regret and sublinear constraint violation bounds for this bandit optimization algorithm in terms of the delays. Finally, numerical results for online quadratically constrained quadratic programs are presented to corroborate the efficacy of the proposed algorithms.

Ship Target Recognition Under Different Sunlight Intensity

<p indent=0mm>In the case of surface target monitoring, the clarity of the ship target often varies with the reflection intensity of the sea surface under different sunlight intensity, which will lead to the unstable recognition rate of the ship target and increase the false alarm rate. For this reason, the ship target recognition algorithm based on ResNet-50 is proposed. Firstly, it uses ResNet-50 network to extract image feature information and applies sunlight robust loss constraint to the features before and after sunlight intensity change to reduce the feature difference. Then, it uses gray-scale histogram to calculate the statistical matrices of features to obtain six features: light contrast, brightness, smoothness, information, third-order matrices and entropy, and generates new feature vector to apply sunlight robust loss constraint to the features before and after sunlight intensity change again. Finally, the two constraints are combined to form a loss function and trained to optimize the optimal weights using Bayesian adaptive hyperparameters. The experimental results show that the average recognition rate of the database for ship sunlight variation reaches 90.47%, which is about 4.00% higher than that before the improvement, and the recognition rate of ship images with sunlight variation of 25, 65 and 80 increases by 3.14%, 6.07% and 16.41%, which shows that the algorithm has a good constraint effect on sunlight variation and the recognition rate is significantly improved.

A Partial Differential Equation-Based Image Restoration Method in Environmental Art Design

With the rapid development of networks and the emergence of various devices, images have become the main form of information transmission in real life. Image restoration, as an important branch of image processing, can be applied to real-life situations such as pixel loss in image transmission or network prone to packet loss. However, existing image restoration algorithms have disadvantages such as fuzzy restoration effect and slow speed; to solve such problems, this paper adopts a dual discriminator model based on generative adversarial networks, which effectively improves the restoration accuracy by adding local discriminators to track the information of local missing regions of images. However, the model is not optimistic in generating reasonable semantic information, and for this reason, a partial differential equation-based image restoration model is proposed. A classifier and a feature extraction network are added to the dual discriminator model to provide category, style, and content loss constraints to the generative network, respectively. To address the training instability problem of discriminator design, spectral normalization is introduced to the discriminator design. Extensive experiments are conducted on a data dataset of partial differential equations, and the results show that the partial differential equation-based image restoration model provides significant improvements in image restoration over previous methods and that image restoration techniques are exceptionally important in the application of environmental art design.

Constraint Loss for Rotated Object Detection in Remote Sensing Images

Rotated object detection is an extension of object detection that uses an oriented bounding box instead of a general horizontal bounding box to define the object position. It is widely used in remote sensing images, scene text, and license plate recognition. The existing rotated object detection methods usually add an angle prediction channel in the bounding box prediction branch, and smooth L1 loss is used as the regression loss function. However, we argue that smooth L1 loss causes a sudden change in loss and slow convergence due to the angle solving mechanism of open CV (the angle between the horizontal line and the first side of the bounding box in the counter-clockwise direction is defined as the rotation angle), and this problem exists in most existing regression loss functions. To solve the above problems, we propose a decoupling modulation mechanism to overcome the problem of sudden changes in loss. On this basis, we also proposed a constraint mechanism, the purpose of which is to accelerate the convergence of the network and ensure optimization toward the ideal direction. In addition, the proposed decoupling modulation mechanism and constraint mechanism can be integrated into the popular regression loss function individually or together, which further improves the performance of the model and makes the model converge faster. The experimental results show that our method achieves 75.2% performance on the aerial image dataset DOTA (OBB task), and saves more than 30% of computing resources. The method also achieves a state-of-the-art performance in HRSC2016, and saved more than 40% of computing resources, which confirms the applicability of the approach.

Exploring the use of knobology for image optimisation in final year sonography students.

Image optimisation is essential for acquisition of quality images in ultrasound and critical to the diagnostic ability of the examination. These skills are taught to sonography students early in their education, but research has found that retention of non-rehearsed knowledge decreases significantly after a year. The aim of this study was to determine which optimisation tools (knobology) final year sonography students use, how often and why they chose to adjust parameters and assess barriers to optimisation of knobology tools. A prospective study using data from an anonymous online survey of 34 final year sonography students. Survey results showed that 19/34 (55%) of students "frequently" optimise all Doppler settings and 23/34 (67%) of students "frequently" optimise basic parameter settings (depth, focus, time gain compensation). Time constraints (24/34 (70%)) and loss of gained knowledge of sonography principles and instrumentation (17/34 (50%)) were the major barriers to the use of knobology. The majority 28/34 (82%) believed they would benefit from further training. This study demonstrates that although most students are optimising settings to improve image quality, sonography principles and instrumentation knowledge loss and time constraints prevent students from maximising machine capabilities. This study supports the need for further training prior to final year clinical placement.

A multi-head neural network with unsymmetrical constraints for remaining useful life prediction

This paper proposes a multi-head neural network (MHNN) model with unsymmetrical constraints for remaining useful life (RUL) prediction of industrial equipment. Generally, the existing deep learning methods proposed for RUL prediction utilize symmetrical constraint loss functions such as the mean squared error function to calculate training errors. However, if the predicted RUL is much larger than the actual value in some safety–critical applications, severe damage may occur. To address this issue, an unsymmetrical constraint function is proposed as the loss function in this work that penalizes the late predictions (i.e., the predicted RUL is larger than the actual RUL) more strongly. In addition, an adjustable parameter is added to this function to adjust the model’s attention to the late predictions. In MHNN model, the bidirectional gated recurrent units (BGRU) and self-attention mechanism are employed to extract temporal features from the condition monitoring data. In addition, the structure of the multi-head neural network is adopted in the proposed model, helping to capture more degradation information by means of multiple identical and parallel networks. The proposed method is validated against a commonly used turbofan engine dataset. Compared with other latest methods on the same dataset, the proposed method is proven to be superior. Taking the FD004 dataset as an example, the score obtained by MHNN is 24.09% lower than that obtained by the best existing method.

Spectral mapping with adversarial learning for unsupervised hyperspectral change detection

Unlike the existing change detection approaches based on the multispectral (MS) image and synthetic aperture radar (SAR) image datasets, a novel unsupervised hyperspectral change detection (UHCD) framework is proposed in this paper. The UHCD framework is designed for hyperspectral images with high dimensions and low availability. This framework consists of two modules: the spectral mapping with adversarial learning and the discriminant analysis with spatial attribute optimization. In comparison with other advanced change detection methods, the proposed framework possesses three distinctive properties: (1) The unsupervised spectral mapping is leveraged to exploit underlying spectral features without the requirement of pseudo-training datasets in the change detection task; (2) We introduce spectral constraint loss into reconstruction space and adversarial loss into latent space to enhance the quality of the features extracted by the spectral mapping network; (3) Spatial attribute optimization uses the spatial correlation to further improve the performance of the proposed UHCD method. The experimental results on two real datasets show that the proposed UHCD achieves competitive performance.

Effective vaccine allocation strategies, balancing economy with infection control against COVID-19 in Japan

Due to COVID-19, many countries including Japan have implemented a suspension of economic activities for infection control. It has contributed to reduce the transmission of COVID-19 but caused severe economic losses. Today, several promising vaccines have been developed and are already being distributed in some countries. Therefore, we evaluated various vaccine and intensive countermeasure strategies with constraint of economic loss using SEIR model to obtain knowledge of how to balance economy with infection control in Japan. Our main results were that the vaccination strategy that prioritized younger generation was better in terms of deaths when a linear relationship between lockdown intensity and acceptable economic loss was assumed. On the other hand, when a non-linearity relationship was introduced, implying that the strong lockdown with small economic loss was possible, the old first strategies were best in the settings of small basic reproduction number. These results indicated a high potential of remote work when prioritizing vaccination for the old generation. When focusing on only the old first strategies as the Japanese government has decided to do, the strategy vaccinating the young next to the old was superior to the others when a non-linear relationship was assumed due to sufficient reduction of contact with small economic loss.

Low-light Image Enhancement Based on Conditional Generative Adversarial Network

We present an end-to-end low-light image enhancement learning method. This learning is based on the conditional generative adversarial networks(GAN) and realizes low-light images enhancement. Specifically, our method uses a convolutional neural network containing residual structures as a generator and WGAN-GP as a discriminator to generate an effective low-light enhancement model under the constraints of GAN loss, Perceptual loss and Structural similarity loss. The model can retain the detailed information of the original image, improve the brightness of the image without generating noise interference, while the generated images are more natural and have higher quality. Extensive experimental results show that our method has reached the state-of-art in multiple objective evaluation indicators of image quality, and the visual appearance is superior.

Continuous Cooling Crystallization in a Coiled Flow Inverter Crystallizer Technology—Design, Characterization, and Hurdles

Continuous small-scale production is currently of utmost interest for fine chemicals and pharmaceuticals. For this purpose, equipment and process concepts in consideration of the hurdles for solids handling are required to transfer conventional batch processing to continuous operation. Based on empirical equations, pressure loss constraints, and an expandable modular system, a coiled flow inverter (CFI) crystallizer with an inner diameter of 1.6 mm was designed. It was characterized concerning its residence time behavior, tested for operation with seed crystals or an ultrasonic seed crystal unit, and evaluated for different purging mechanisms for stable operation. The residence time behavior in the CFI corresponds to ideal plug flow behavior. Crystal growth using seed crystals was demonstrated in the CFI for two amino acids. For fewer seed crystals, higher crystal growth rates were determined, while at the same time, secondary nucleation was observed. Feasibility for the interconnection of a sonicated seeding crystal unit could be shown. However, the hurdles are also identified and discussed. Prophylactic flushing combined with a photosensor for distinguishing between solvent and suspension phase can lead to stable and resource-efficient operation. The small-scale CFI technology was investigated in detail, and the limits and opportunities of the technology are presented here.

Decoupling Analysis of Interaction between Tunnel Surrounding Rock and Support in Xigeda Formation Strata

The Xigeda Formation has the characteristics of poor cementation, variable structure, and softening in water. It is prone to lead to failure of the initial support structures, arch collapse, and even roof fall. The key to the problem lies in the exploration of the interaction between surrounding rock and support. Therefore, based on the constraint loss theory of the tunnel face space effect, this paper constructively introduces multiple dependent variables, including the intermediate principal stress, the dilatancy of the surrounding rock, the timeliness of the shotcrete support stiffness, and the water content of the Xigeda formation, to decouple the whole process of the interaction between the surrounding rock and support, and discusses the influence of each variable. The research results show that: i) there is a critical support point in the tunnel under the influence of the unified strength theoretical parameter b and water content, and when the unified strength theoretical parameter b is constant, the critical distance of the support will decrease with the increase of the water content; ii) from the perspective of deformation control, the stress of support structure will increase with the increase of dilatancy angle of Xigeda surrounding rock; iii) considering the hardening characteristics of shotcrete, the initial support stiffness of Xigeda tunnel increases nonlinearly, and the whole process of stress and deformation can be divided into four stages; iv) the combined support structure of section steel and grille has little difference in the deformation control effect of the surrounding rock and the stress of the support structure, but the combined support structure of the grille steel frame is more sensitive to the hardening parameters.

A Diachronic Change of Negation Variation in Spoken Korean: from the 17&lt;SUP&gt;th&lt;/SUP&gt; century to the 2010s

This study aims to explore how negation variation between short-negation (S-Neg) and long-negation (L-Neg) has changed in spoken Korean. Concerning the topic, two research questions were addressed: 1) how negation variation has changed in spoken Korean from the 17^th century to the 2010s? 2) how can the change be explained linguistically? To answer these questions, 1,894 negative sentences were collected from oral literature and drama scripts between the 17^th century and the 2010s and the proportion of L-Neg in the data was calculated along the time. The results revealed that 1) L-Neg has never spread widely in the Kyeongsang/Cheolla dialects while it had prevailed in the Seoul dialect and has declined until the 2010s; 2) A morpho-syntactic constraint of S-Neg, which was claimed to result in the spread of L-Neg in Middle Korean (Park, 2005, 2011) was lost in contemporary Korean. Based on the findings, I suggest that the decline of L-Neg in spoken Korean is ascribed to the loss of the S-Neg constraint and propose a morpho-syntactic change of the negative word an(i) in S-Neg.