Search

In recent years, the proximal policy optimization (PPO) algorithm has received considerable attention because of its excellent performance in many challenging tasks. However, there is still a large space for theoretical explanation of the mechanism of PPO's horizontal clipping operation, which is a key means to improve the performance of PPO. In addition, while PPO is inspired by the learning theory of trust region policy optimization (TRPO), the theoretical connection between PPO's clipping operation and TRPO's trust region constraint has not been well studied. In this article, we first analyze the effect of PPO's clipping operation on the objective function of conservative policy iteration, and strictly give the theoretical relationship between PPO and TRPO. Then, a novel first-order policy gradient algorithm called authentic boundary PPO (ABPPO) is proposed, which is based on the authentic boundary setting rule. To ensure the difference between the new and old policies is better kept within the clipping range, by borrowing the idea of ABPPO, we proposed two novel improved PPO algorithms called rollback mechanism-based ABPPO (RMABPPO) and penalized point policy difference-based ABPPO (P3DABPPO), which are based on the ideas of rollback clipping and penalized point policy difference, respectively. Experiments on the continuous robotic control tasks implemented in MuJoCo show that our proposed improved PPO algorithms can effectively improve the learning stability and accelerate the learning speed compared with the original PPO.

Adaptive gradient algorithms have recently become extremely popular because they have been applied successfully in training deep neural networks, such as Adam, AMSGrad, and AdaBound. Despite their success, however, the distributed variant of the adaptive method, which is expected to possess a rapid training speed at the beginning and a good generalization capacity at the end, is rarely studied. To fill the gap, a distributed adaptive subgradient algorithm is presented, called D-AdaBound, where the learning rates are dynamically bounded by clipping the learning rates. Moreover, we obtain the regret bound of D-AdaBound, in which the objective functions are convex. Finally, we confirm the effectiveness of D-AdaBound by simulation experiments on different datasets. The results show the performance improvement of D-AdaBound relative to existing distributed online learning algorithms.

Real-time rendering of pixel-accurate hard shadows in 3D scenes is still a challenging problem for shadow generation techniques. In this paper, we present the concept of non-overlapping geometric shadow map (NOGSM), which enables the construction of high-quality hard shadows from static lights for any point in the scene. The NOGSM is created in a preprocessing phase, in which the scene’s geometry is looked at from the light’s point of view. All hidden surfaces are removed through total elimination of hidden polygons plus clipping of the partially hidden polygons. The resulting list of non-overlapping projected polygons is then stored in a multi-grid data structure. Our solution generates shadows with far better quality than standard shadow mapping techniques, while maintaining real time performance, and using a smaller memory footprint than compression techniques.

To solve the problem of ergonomic evaluation of suit production system, a Hybrid method with IAHP and gray entropy theory is proposed. Firstly, literature review about ergonomics evaluation tools and technology is introduced. Furthermore, the ergonomic evaluation index system of suit operation is established according to production practice. Besides, comprehensive evaluation model process with gray entropy and IAHP is introduced. Finally, detailed evaluation process of suit sewing, ironing and clipping operation is introduced by an example, and the results show that the evaluation method in the paper is helpful to verify the feasibility and effectiveness of the operation system improvement demonstration.

Micro aerial vehicles (MAVs) with multiple rotors, or multicopters, have many promising applications ranging from environmental monitoring, agricultural inspection, to package delivery. These applications, however, usually face a critical problem: the flight time of MAVs is limited due to the low aerodynamic efficiency and high energy consumption. One promising solution is to make them rest on desired objects using perching, an important capability in biological flyers (e.g., birds). In this article, we present the design and experimentation of a novel perching mechanism: a low cost, 3-D-printed gripper with bistability (i.e., two stable states). The gripper has two unique characteristics. First, using bistability, it can passively switch from open to closed state using the impact between the gripper and the perching object, alleviating the requirement for precise motion control. Second, the gripper has two perching methods for different objects. For objects with a small height, it can form a closed diamond shape to encircle the objects (encircling method). For objects with a large height, the gripper's two fingers can clip on each side of the objects to utilize the friction forces for perching (clipping method). We analyze the proposed gripper design to predict the required force for opening and closing the gripper. We also predict the size of objects that will allow for successful perching for the clipping method. All the theoretical analyses are experimentally verified. Finally, we integrate the gripper onto a palm-size quadcopter to enable a mechatronic system for perching, and demonstrate successful perching with both clipping and encircling methods as well as aerial grasping. Although our bistable gripper is used with a palm-size quadcopter, the design strategy can also be applied to large-size MAVs for both energy efficient perching and aerial grasping.

The growing popularity of convolutional neural networks (CNNs) has led to the search for efficient computational platforms to accelerate CNN training. Resistive random-access memory (ReRAM)-based manycore architectures offer a promising alternative to commonly used GPU-based platforms for training CNNs. However, due to the immature fabrication process and limited write endurance, ReRAMs suffer from different types of faults. This makes training of CNNs challenging as weights are misrepresented when they are mapped to faulty ReRAM cells. This results in unstable training, leading to unacceptably low accuracy for the trained model. Due to the distributed nature of the mapping of the individual bits of a weight to different ReRAM cells, faulty weights often lead to exploding gradients. This in turn introduces a positive feedback in the training loop, resulting in extremely large and unstable weights. In this paper, we propose a lightweight and reliable CNN training methodology using weight clipping to prevent this phenomenon and enable training even in the presence of many faults. Weight clipping prevents large weights from destabilizing CNN training and provides the backpropagation algorithm with the opportunity to compensate for the weights mapped to faulty cells. The proposed methodology achieves near-GPU accuracy without introducing significant area or performance overheads. Experimental evaluation indicates that weight clipping enables the successful training of CNNs in the presence of faults, while also reducing training time by 4 X on average compared to a conventional GPU platform. Moreover, we also demonstrate that weight clipping outperforms a recently proposed error correction code (ECC)-based method when training is carried out using faulty ReRAMs.

Nowadays, field-programmable gate array (FPGA) devices have been widely used in various fields. However, modules of circuits to be executed on FPGAs are placed within rectangular reconfigurable regions (RRs) with current floorplanners, leading to internal fragments, and lower utilization of resources. To address this, a dynamic description model of RRs and the corresponding floorplanner named dynamic union partial reconfiguration floorplan (DUPRFloor) are proposed in this article. The RR dynamic description modeling adds an anchor within a rectangular RR to reduce internal fragments. In this way, the modeling can represent both rectangular and nonrectangular shapes. Then, to find the optimal anchor, a clipping method is devised by constraining width and height of the candidate region. Finally, the mixed-integer linear programming (MILP) is used to optimize an objective function which considers the resources utilization and communication costs to obtain a desirable floorplanning result. The proposed method has been validated by simulation on three kinds of devices. And experimental results show that reconfigurable resources can be saved as much as 19.16% compared to rectangular modeling method. The DUPRFloor is also validated on the Microelectronics Center of North Carolina standard benchmark data sets. Results show that DUPRFloor can reduce 18.65% global wire length at most with almost the same execution time compared to state-of-the-art algorithms. Our approach is tested on a FPGA implemented software-defined radio (SDR) and reduced 29.41% wasted configurable frames, and to the overall design, 2% configurable frames are saved at most.

This article proposes a demand-side management (DSM) mechanism for energy management based on user behavior monitoring in a smart home. In the proposed mechanism, first through an analytic hierarchy process, the most influential factors related to power consumption are extracted. Next, by employing the K-means algorithm on the extracted factors, users are clustered. The user's clusters, the power grid state, and the user's real-time power consumption are inputs for a control unit. We present an interactive algorithm for the control unit, which causes peak reduction using peak clipping techniques. We also develop a day-ahead scheduling mechanism, which optimizes the load based on load shifting techniques. The proposed system is implemented in an Internet of Things (IoT) testbed consisting of four tiers—sensors, home gateways, server, and web portal. The central server is based on the Kaa IoT platform, an open-source platform widely used in the IoT domain. The performance of the proposed system is evaluated through simulation and a case study. Results confirm that the proposed system reduces the power consumption and costs for users and improves power grid performance in terms of the peak-to-average ratio.

Unmanned autonomous helicopter (UAH) path planning problem is an important component of the UAH mission planning system. The performance of the automatic path planner determines the quality of the UAH flight path. Aiming to produce a high-quality flight path, a path planning system is designed based on human-computer hybrid augmented intelligence framework for the UAH in this paper. Firstly, an improved artificial bee colony (I-ABC) algorithm is proposed based on the dynamic evaluation selection strategy and the complex optimization method. In the I-ABC algorithm, the following way of on-looker bees and the update strategy of nectar source are optimized to accelerate the convergence rate and retain the exploration ability of the population. In addition, a space clipping operation is proposed based on the attention mechanism for constructing a new spatial search area. The search time can be further reduced by the space clipping operation under the path planning result within acceptable changes. Moreover, the entire optimization process and results can be feeded back to the knowledge database by the human-computer hybrid augmented intelligence framework to guide subsequent path planning issues. Finally, the simulation results confirm that a feasible and effective flight path can be quickly generated by the UAH path planning system based on human-computer hybrid augmented intelligence.

Proximal policy optimization (PPO) has yielded state-of-the-art results in policy search, a subfield of reinforcement learning, with one of its key points being the use of a surrogate objective function to restrict the step size at each policy update. Although such restriction is helpful, the algorithm still suffers from performance instability and optimization inefficiency from the sudden flattening of the curve. To address this issue we present a novel functional clipping policy optimization algorithm, named Proximal Policy Optimization Smoothed Algorithm (PPOS), and its critical improvement is the use of a functional clipping method instead of a flat clipping method. We compare our approach with PPO and PPORB, which adopts a rollback clipping method, and prove that our approach can conduct more accurate updates than other PPO methods. We show that it outperforms the latest PPO variants on both performance and stability in challenging continuous control tasks. Moreover, we provide an instructive guideline for tuning the main hyperparameter in our algorithm.

Algo que foi mapeado possui, em primazia, o indicativo que seus caminhos constitutivos são plenos em conhecimento e encontram-se sob a forma de registros de qualquer natureza. O ato de mapear requer dedicação na realização dos registros, podendo servir para que outrem seja capaz de seguir seus caminhos em segurança ou perder-se mais bem orientado. O texto aborda o processo tecnológico de recorte da luz, enquanto objeto da Arte, sobre uma superfície irregular. Mapeiam-se algumas experiências da ação de dobrar a imagem em movimento sobre diferentes arquiteturas e objetos pela projeção mapeada. Assim, refletindo a respeito do trajeto entre composição audiovisual e técnica sensível à superfície, descreve-se o processo de compartilhar essa experiência audiovisual com outros artistas, por meio de uma dinâmica de encontros acadêmicos, objetivando a realização de um evento coletivo com contornos e singularidades de forma limítrofe na concepção e realização em tais produções.

This is a proof-of-concept study to create a four-dimensional (4-D) cine model of the heart and visualize it in virtual reality by using freely available open-source software and inexpensive hardware. Four-dimensional cine models allow for real-time visualization of cardiac structures during processes such as complex congenital heart disease. Such models can be used for patient and trainee education, and potentially for surgical planning. Currently, 3-D printed models are more commonly used, but they are static, showing only one selected phase of the cardiac cycle. Second, they are limited by the selection of clipping planes before printing. Four-dimensional segmentation and virtual reality visualization overcome these limitations. Currently, most of the work in virtual/augmented reality models involves the segmentation of one cardiac phase or the use of expensive software for multiphase segmentation. In this study, we show an approach for multiphase cardiac segmentation as well as its display using free open-source software and relatively inexpensive hardware.

The most commonly used and transferred files are audio files. However, when transferring such files, we face the problem of detecting any modifications. The widespread use of sound editors and various programmes for the processing and editing of audio files makes it possible for anyone to falsify speech phonograms. Another type of modification involves using audio files as steganographic containers. There are a large number of steganography techniques for the covert transmission of information within standard files. In this regard, it is important to develop analytical tools and methods for the detection of modifications in audio files. There fore, the article presents a study on the detection of hidden steganographic messages and signs of modifications in the data area of audio files based on determining the numerical criteria for decision making. The article considers methods for the statistical analysis of audio files for any modifications in the data. When analysing the alterations made to audio files, steganalysis methods and spectral analysis were used. The article presents algorithms based on the assessment of changes in the statistical characteristics of the signal-to-noise ratio, which make it possible to evaluate the modifications. Algorithms for calculating the coefficient of the signal-to-noise ratio are described both for separate blocks and for the whole file. The implemented algorithms were tested on audio files, which were preliminary modified in the following ways: embedding — using the audio file as a stegocontainer; merging - using two audio files to make one; clipping — modifying the audio file by deleting some of the information. Using the proposed methods, we evaluated the possibility of modifications made to the files. The experiment demonstrated that the presented methods allow the detection of merged audio files and removed fragments without using the original. The article provides data that shows the accuracy of modification detection.

This letter presents a novel algorithm for Sensor-to-Device Line of Sight (LoS) detection for point cloud representations of indoor scenarios. Perspective projection with region clipping is utilized to evaluate the relative position of an obstacle coordinate cloud with respect to the direct sensor-device visibility. Performance of the proposed algorithm is verified against the classical ray-voxel intersection algorithm. The letter concludes with a description of the algorithm’s applicability and limitations.

Herein we propose a new approach for deducing the topology of metal–organic frameworks (MOFs) assembled from organic ligands of low symmetry, which we call net-clipping. It is based on the construction of nets by rational deconstruction of edge-transitive nets comprising higher-connected molecular building blocks (MBBs). We have applied net-clipping to predict the topologies of MOFs containing zigzag ligands. To this end, we derived 2-connected (2-c) zigzag ligands from 4-c square-like MBBs by first splitting the 4-c nodes into two 3-c nodes and then clipping their two diagonally connecting groups. We demonstrate that, when this approach is applied to the 17 edge-transitive nets containing square-like 4-c MBBs, net-clipping leads to generation of 10 nets with different underlying topologies. Moreover, we report that literature and experimental research corroborate successful implementation of our approach. As proof-of-concept, we employed net-clipping to form three new MOFs built with zigzag ligands, each of which exhibits the deduced topology., This work was supported by the Spanish MINECO (project RTI2018-095622-B-I00), the Catalan AGAUR (project 2017 SGR 238), and the ERC under the EU-FP7 (ERC-Co 615954). It was also funded by the CERCA Program/Generalitat de Catalunya. ICN2 is supported by the Severo Ochoa program from the Spanish MINECO (Grant No. SEV-2017-0706).

Aimed at current problems in tour route planning, this research proposes a tour route planning algorithm based on the data mining of tourist sights based on their precise interest. It further studies on the aspect of tourist interests and needs, tourist sight feature attributes, data mining of tourist sights based on their precise interest and optimal tour route planning, and then designs and performs experiment to testify the feasibility and superiority of the algorithm. First, tourist interest and need label matrix and tourist sight feature attribute matrix are set up. Tourist sight attribute factor is formed by mining text encyclopedic data, and combining with other factors, tourist sight clustering algorithm is set up to output tourist sight clusters. Second, data mining algorithm of tourist sights based on their precise interest is set up, which outputs the optimal tourist sights not only match the tourists' interests and needs but also optimally distribute around to produce the lowest travel costs. Third, based on precise interested tourist sights, combining with geographic information data, traffic information data and tourist sight information data, optimal tour route planning algorithm based on interest label is set up. It outputs maximum heap and complete binary tree with descending order interval motive values and corresponding edge clipping circle, which directly reflects the satisfaction content of each tour route on tourists. Tourist sight domain and geographic spatial data of Zhengzhou city are collected as basic data to design the experiment, and it testifies the algorithm's feasibility, effectiveness and practicalness. Meanwhile, the experiment brings in three commonly used route planning algorithms as control group to make comparison with the developed algorithm on the aspect of optimal route, motive value, motive value difference, algorithm performance and guide map,etc., which testifies that the developed algorithm has superiority to traditional algorithms. The research indicates that this developed algorithm solves the problems on tour route planning and it is an effective, feasible and practical algorithm.

The paper considers a single-pass scheme for rendering dynamic 3D scenes using modern GPUs and graphics interfaces. The following methods and techniques are used with this scheme: clipping of objects using spatial decomposition and indexing methods, hardware occlusion queries, fragmentation and caching of command buffers. These methods require significant computational resources to execute, and the amount of work in the stages of the graphics pipeline depends on their results. Therefore, a balanced use of resources when transferring graphics data and executing commands is important. A performance model of the graphics pipeline is proposed, which makes it possible to estimate the required resources depending on the applied base methods and characteristics of the displayed scene. Unlike existing methods and models, the proposed model allows to calculate the costs of composing command buffers using various recording techniques, the costs of sending, executing, and receiving the results of hardware occlusion queries. Formulas are derived to calculate frame rendering time depending on the number of occlusion queries. A method is proposed to estimate the number of occlusion queries for efficient rendering of dynamic scenes. Computational experiments are carried out to show the relevance of the proposed model and the effectiveness of the developed method when displaying large dynamic scenes. Section 1 provides an overview of related work as well as general purposes of given paper and its structure. Section 2 describes the proposed performance model and method used to calculate the number of occlusion queries for efficient rendering of dynamic scenes. Section 3 presents the performance analysis, which contains derived and measured rendering time when rendering scenes and employing frustum culling, occlusion queries. Section 4 summarizes the main conclusions.

Background and aim Although many algorithms have been proposed to segment brain structures in MRI scans, comparison of different algorithms in the same data set is rarely done. Many methods still run on privately held data and include comparisons with previous versions. The purpose of this study is to introduce a new Generative Adversarial Network (GAN) based segmentation architecture. Brain tissues on 3D-MRI scans with T1w modality were segmented into three (GM, WM, CSF) and eight (CGM, BG, WM, WMH, CF, VE, CE ve BS) parts. Methods The proposed approach (Vol2SegGAN) consists of two parts: preprocessing and segmentation. The preprocessing includes extraction of the brain region, editing of labels in the datasets, MNI152 registration, clipping/sampling. The segmentation part is carried out by the collaboration of a generator (includes ACFP and PAM modules) and a discriminator (distinguishes real/fake) architectures. Conclusions In the three part segmentation process, the proposed method showed the best segmentation success in CSF (Dice = 0.739, VS = 0.967), GM (Dice = 0.878, HD = 2.378) and WM (HD = 2.105) tissues, and the second best segmentation success in WM (Dice = 0.793, VS = 0.972) tissue according to Dice and VS metrics. Similarly, in the segmentation of eight parts, it is seen that it has the best success according to VS and HD metrics and the second best according to Dice metric. Vol2SegGAN, which has fewer parameters (6,883 mil.) than existing architectures, has an average of 11–12 s (CPU) or 0–1 s (GPU) to segment a sample 3D-MRI. Implementation codes of the proposed architecture are available on the github page 1 .

Augmented reality (AR) is an emerging technology to improve visualization experiences. However, visualizing volume data is limited in existing AR systems due to the lack of an intuitive and precise exploration scheme. In this paper, we present a 3D augmented volume visualization and screen-based interaction method. An autostereoscopic handheld display is utilized to achieve naked-eye 3D perception, and a stereo camera is adopted to track the display’s 6 DoF pose. We implement real-time ray casting with GPU acceleration and enhance the visual experience by defining the dynamic view frustum, clipping interaction and transfer function based on the pose of the handheld display. Our display system achieves real-time rendering and tracking performance with dynamic visual effects, allowing a global overview and detailed visualization of arbitrary clipping planes. We also perform a user study to compare an anatomical landmark annotation task on a 2D environment with our system. The results show a significant reduction in completion time and an improvement in depth perception and comprehension of complex topologies. Furthermore, to illustrate the applicability of our system, we present three volumes from different biological scales.

Temporal Action Detection is a research hotspot in Video Understanding. It is necessary not only to give the specific moments of the beginning and end of each action instance in the video, but also to give the category of the action instance. At present, most of the methods in temporal action detection are divided into two steps: The video is divided into a series of video clips to determine whether each clip is an action instance, and the fragments that are not an action instance are deleted; Segments that may be action instances are then classified to get the final result. At present, the difficulties in action detection research mainly include the following two points: 1) The boundary is not clear. Different from action recognition, action detection requires precise positioning, but an action in life is often not very definite. It is also the reason why the mean average precision (mAP) of action detection is low at present. 2) The time span is large. In life, an action often spans a very long scale. Short movements such as waving can last for a few seconds, while long movements such as rock climbing or cycling can last for tens of minutes, which makes it extremely difficult for us to extract proposals. In view of the above difficulties, this paper proposes a large receptive field boundary matching network(LRFBMN) model which takes advantage of the relationship between proposals to improve the accuracy of proposal generation. The model is mainly divided into two parts: 1) Clipping feature map is processed by large kernel convolution, and then proposal feature is generated by ROI-pooling; 2) The proposals are arranged in a certain order to form a fixed graph, and the information exchange between graph nodes is realized by using convolution with dilation. Through experiments, this model is 2.35% higher than baseline and 1.06% higher than state-of-the-art in THUMOS14 data set.

Online Learning to Rank (OLTR) has been primarily studied in the centralised setting, where a central server is responsible to index the searchable data, collect the users' queries and search interactions, and optimize ranking models. A drawback of such a centralised OLTR paradigm is that it cannot guarantee user's privacy as all data (both the searchable one and the one related to user interactions) is collected by the server. In this paper, we propose a Federated OLTR method, called FPDGD, which leverages the state-of-the-art Pairwise Differentiable Gradient Descent (PDGD) and adapts it to the Federated Averaging framework. For a strong privacy guarantee, we further introduce a noise-adding clipping technique based on the theory of differential privacy to be used in combination with FPDGD. Empirical evaluation shows FPDGD significantly outperforms the only other federated OLTR method. In addition, FPDGD is more robust across different privacy guarantee requirements than the current method: our method is therefore more reliable for real-life applications.

In organizations, companies or agencies such as the Office of Communication and Information Technology, presence has become one of the critical things in assessing the performance of workers/employees. There are several ways to do attendance, where one of them can facilitate the present process with digital tools/machines such as smartphones that utilize the GPS feature to run APIs such as Google Geolocation on Android. Also, a system is needed that can manage the results of the present process, such as web-based applications that can also be accessed through either a computer or smartphone. Therefore, the need to build a presence application both in the presence and in management that can help agencies such as Office of Communication and Information (Diskominfo) to handle internal processes, especially in the management of its employees. In this study, the application in the construction of its application is used Point Clipping algorithm as a tool for the present process that can take advantage of the GPS feature. Based on the results of the tests conducted, the accuracy value of the incoming and returning attendance process was obtained in the location area with an average of each side of the area of approximately 11.21 meters at the entry point and 18.05 meters at the presence of going home outside the office, where the location area boundary outside the specified office which is around 15 meters and the time test is obtained the time delay value in the attendance process with an average of each side of the area of about 9.6 seconds at the incoming attendance and 8.3 seconds at the home attendance.

We present a framework for time-varying 3-D ultrasonic testing datasets based on real-time CUDA-based volume rendering with high quality. In addition, we design novel trapezoid-shaped transfer functions with a color gradient. Furthermore, we propose an interactive 3-D inspection method via clipping plane with enhanced ability of the cross-section data.

The present paper deals with a specific parameter estimation problem. Assuming a time series of 0-1-valued variables {Xn, n∈N0} obtained by clipping a Gaussian AR(p) time series {Zn, n∈N0} at unkno...

The root-finding problem has wide applications in geometric processing and computer graphics. Previous rational cubic clipping methods are either of convergence rate 7/k with O(n2) complexity, or of convergence rate 5/k with linear complexity, where n and k are degree of the given polynomial and the multiplicity of the root, respectively. This paper presents an improved rational cubic clipping of linear complexity, which can achieve convergence rate 7/k, or a better one 7 / ( k − 1 ) for a multiple root such that k ≥ 2. Numerical examples illustrate both efficiency and convergence rate of the new method.

We consider an iterative compensation method to treat clipping distortion in orthogonal frequency division multiplexing systems with clipping. A conventional approach is using extrinsic information (EI) from a decoder to estimate and cancel clipping distortion iteratively. In this letter, feedback in the form of a posteriori information (API) from the decoder is investigated. Correlation may be introduced among messages in iterative processing when API is directly used as feedback. We show that carefully choosing a linearization model can suppress such correlation. Simulation results show that the API approach outperforms the EI one.

The visualization toolkit (VTK) is a popular cross-platform, open source toolkit for scientific and medical data visualization, processing, and analysis. It supports a wide variety of data formats, algorithms, and rendering techniques for both polygonal and volumetric data. In particular, VTK's volume rendering module has long provided a comprehensive set of features such as plane clipping, color and opacity transfer functions, lighting, and other controls needed for visualization. However, due to VTK's legacy OpenGL backend and its reliance on a deprecated API, the system did not take advantage of the latest improvements in graphics hardware or the flexibility of a programmable pipeline. Additionally, this dependence on an antiquated pipeline posed restrictions when running on emerging computing platforms, thereby limiting its overall applicability. In response to these shortcomings, the VTK community developed a new and improved volume rendering module, which not only provides a modern graphics processing unit-based implementation, but also augments its capabilities with new features such as fast volume clipping, gradient-magnitude-based opacity modulation, render to texture, and hardware-based volume picking.

This paper proposes formulating declipping as a constrained multiple measurement vector (MMV) optimization problem that has a ${\ell _{2,0}}$ group norm as its cost function for further improving the state-of-the-art declipping method SParse Audio DEclipper (SPADE). This paper shows that the MMV optimization problem can be solved by extending the steps of alternating direction method of multipliers (ADMM) in the SPADE algorithm. The proposed method improved the signal-to-distortion ratio and reduced the cepstral distance for all clipping levels in a numerical simulation.

This paper presents a convolutional neural network (CNN) accelerator that can skip zero weights and handle outliers, which are few but have a significant impact on the accuracy of CNNs, to achieve speedup and increase the energy efficiency of CNN. We propose an offline weight-scheduling algorithm which can skip zero weights and combine two non-outlier weights simultaneously using bit-level sparsity of CNNs. We use a reconfigurable multiplier-and-accumulator (MAC) unit for two purposes; usually used to compute combined two non-outliers and sometimes to compute outliers. We further improve the speedup of our accelerator by clipping some of the outliers with negligible accuracy loss. Compared to DaDianNao [7] and Bit-Tactical [16] architectures, our CNN accelerator can improve the speed by 3.34 and 2.31 times higher and reduce energy consumption by 29.3% and 30.2%, respectively.

Forensic craniofacial identification encompasses the practices of forensic facial approximation (aka facial reconstruction) and craniofacial superimposition within the field of forensic art in the United States. Training in forensic facial approximation methods historically has used plaster copies, high-cost commercially molded skulls, and photographs. Despite the increased accessibility of computed tomography (CT) and the numerous studies utilizing CT data to better inform facial approximation methods, 3D CT data have not yet been widely used to produce interactive resources or reference catalogs aimed at forensic art practitioner use or method standardization. There are many free, open-source 3D software packages that allow engagement in immersive studies of the relationships between the craniofacial skeleton and facial features and facilitate collaboration between researchers and practitioners. 3D CT software, in particular, allows the bone and soft tissue to be visualized simultaneously with tools such as transparency, clipping, and volume rendering of underlying tissues, allowing for more accurate analyses of bone to soft tissue relationships. Analyses and visualization of 3D CT data can not only facilitate basic research into facial variation and anatomical relationships relevant for reconstructions but can also lead to improved facial reconstruction guidelines. Further, skull and face surface models exported in digital 3D formats allow for 3D printing of custom reference models and novel training materials and modalities for practitioners. This chapter outlines the 3D resources that can be built from CT data for forensic craniofacial identification methods, including how to view 3D craniofacial CT data and modify surface models for 3D printing.

This paper presents a new approach to line clipping by a convex polygon problem solution. The algorithm is based on a separation function, which separates the polygon vertices to the left or right hand side of the given line. It leads to numerically robust algorithms in comparison to the well-known Cyrus-Beck’s algorithm and its modifications.

PurposeThe treatment of cerebral aneurysms shifted from microsurgical to endovascular therapy. But for some difficult aneurysm configurations, e.g. wide neck aneurysms, microsurgical clipping is better suited. From this combination of limited interventions and the complexity of these cases, the need for improved training possibilities for young neurosurgeons arises.MethodWe designed and implemented a clipping simulation that requires only a monoscopic display, mouse and keyboard. After a virtual craniotomy, the user can apply a clip at the aneurysm which is deformed based on a mass–spring model. Additionally, concepts for visualising distances as well as force were implemented. The distance visualisations aim to enhance spatial relations, improving the navigation of the clip. The force visualisations display the force acting on the vessel surface by the applied clip. The developed concepts include colour maps and visualisations based on rays, single objects and glyphs.ResultsThe concepts were quantitatively evaluated via an online survey and qualitatively evaluated by a neurosurgeon. Regarding force visualisations, a colour map is the most appropriate concept. The necessity of distance visualisations became apparent, as the expert was unable to estimate distances and to properly navigate the clip. The distance rays were the only concept supporting the navigation appropriately.ConclusionThe easily accessible surgical training simulation for aneurysm clipping benefits from a visualisation of distances and simulated forces.

Recent advances in audio declipping have substantially improved the state of the art.% in certain saturation regimes. Yet, practitioners need guidelines to choose a method, and while existing benchmarks have been instrumental in advancing the field, larger-scale experiments are needed to guide such choices. First, we show that the clipping levels in existing small-scale benchmarks are moderate and call for benchmarks with more perceptually significant clipping levels. We then propose a general algorithmic framework for declipping that covers existing and new combinations of variants of state-of-the-art techniques exploiting time-frequency sparsity: synthesis vs. analysis sparsity, with plain or structured sparsity. Finally, we systematically compare these combinations and a selection of state-of-the-art methods. Using a large-scale numerical benchmark and a smaller scale formal listening test, we provide guidelines for various clipping levels, both for speech and various musical genres. The code is made publicly available for the purpose of reproducible research and benchmarking.

Convex hull is one of the basic means to describe the shape of objects, which is used in many fields of computer graphics and images. This paper proposes a support vector machine classification method based on convex hull clipping. Principle component analysis (PCA) is used for pretreatment on the planar point set. The end point of convex hull is determined by the PCA fitting line, then convex hull is obtained by the rapid sorting method. During classification, the approximating line of the boundary is constructed by translating the fitting line to the class boundary. Samples in the region surrounded by the approximating line and convex hull are boundary samples, which are used to the train support vector machine (SVM). Experiments on two artificial data sets and the UCI standard data set show that the proposed method can improve classification accuracy and reduce training time, especially when the training set is large.

To solve the problem brought by high peak-to-average power ratio (PAPR) in orthogonal frequency division multiplexing (OFDM) system, a joint PAPR reduction scheme based on adaptive clipping algorithm and DHT precoding (ACA-DHT) is proposed in this paper. In this scheme, we use a feedback mechanism based on statistical characteristics to improve the clipping threshold setting. When the complementary cumulative distribution function (CCDF) of PAPR is equal to 10-3, the PAPR values of our proposed scheme are 7.76dB, 5.75dB, 5.07dB, 4.04dB, 1.48dB and 0.51dB lower than those of OFDM system, PTS, SLM, clipping, DHT, DHT-clipping, respectively.

Throughout the past several years, the new technologies and enhancements related to medical visualization has been more than fascinating. In this study, we discuss a proposed idea of a handheld or a mobile Augmented Reality (AR) application. This application will ease the process of medical visualization. The criticality of a faulty medical imaging diagnosis could lead to severe consequences and outcomes to patients’ health. This criticality is what ignites new ideas to improve medical imaging. The proposed AR application will help doctors and radiologists in their daily basis. Still most doctors, surgeons, and radiologists use two-dimensional (2D) slices to visualize patients’, which is quite hard to image and map the 3D structure of a human anatomical structure with just 2D plain images. This process requires a lot of experience which sometimes might not be enough. It will help physicians by visualizing patients’ scans such as Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) to be rendered as a three-dimensional (3D) object and not as a 2D object. Using the AR technology makes it much more straightforward in observing the rendered object. The proposed application reads a medical image file(s) such as CT or MRI files and renders its data as a 3D model in real world space, by using a volume rendering technique called ray-tracing. This technique works by reading data from a selected file or dataset. After reading a file, the technique traces a ray into object space from each pixel. Afterwards, it accumulates and calculates the color and opacity values of pixels along each ray. The Final step is to assign the values to its corresponding pixels. By completing this process, a selected dataset is rendered as a 3D object into real world space. With AR technology, seeing and interacting with the 3D object is much easier for the users. The expected outcomes of this study are to improve the process of pre-surgery medical images visualization. With interactive buttons and controllers included in the application and the use of several options like decreasing opacity, clipping, zooming, rotation, threshold … etc. will give more control to the users. Moreover, the proposed application enables users to customize his views to get the best of the 3D object for better visualization. This study discusses the in development mobile AR application that could be used as a new, more facilitating method of visualizing medical images. The application will deliver a more feasible way of interacting with visualized medical images, with buttons and sliders and virtual joysticks creating a better user experience (UX).

With the widespread use of handheld devices, 3D graphics processing capability has become a differentiating factor in mobile SoC design. This paper presents the design of a fully programmable 3D graphics processor for mobile applications. We adopt single instruction, multiple data (SIMD) architecture and multi-thread scheduling to utilize data and instruction parallelism in the unified shader. And in rendering engine design, we propose distributed clipping and hierarchical tile-based rasterization to improve the speed. Hardware simulation results show that the triangle processing speed is improved by 33.3% from six to four clock cycles. And the synthesis results in SMIC 40 nm technology show that the maximum frequency is 600 MHz, and the triangle processing speed is 150 MTri/s.

The availability of commercial fully immersive virtual reality systems allows the proposal and development of new applications that offer novel ways to visualize and interact with multidimensional neuroimaging data. We propose a system for the visualization and interaction with Magnetic Resonance Imaging (MRI) scans in a fully immersive learning environment in virtual reality. The system extracts the different slices from a DICOM file and presents the slices in a 3D environment where the user can display and rotate the MRI scan, and select the clipping plane in all the possible orientations. The 3D environment includes two parts: 1) a cube that displays the MRI scan in 3D and 2) three panels that include the axial, sagittal, and coronal views, where it is possible to directly access a desired slice. In addition, the environment includes a representation of the brain where it is possible to access and browse directly through the slices with the controller. This application can be used both for educational purposes as an immersive learning tool, and by neuroscience researchers as a more convenient way to browse through an MRI scan to better analyze 3D data.

Polyhedral meshes are increasingly becoming an attractive option with particular advantages over traditional meshes for certain applications. What has been missing is a robust polyhedral meshing algorithm that can handle broad classes of domains exhibiting arbitrarily curved boundaries and sharp features. In addition, the power of primal-dual mesh pairs, exemplified by Voronoi-Delaunay meshes, has been recognized as an important ingredient in numerous formulations. The VoroCrust algorithm is the first provably-correct algorithm for conforming polyhedral Voronoi meshing for non-convex and non-manifold domains with guarantees on the quality of both surface and volume elements. A robust refinement process estimates a suitable sizing field that enables the careful placement of Voronoi seeds across the surface circumventing the need for clipping and avoiding its many drawbacks. The algorithm has the flexibility of filling the interior by either structured or random samples, while preserving all sharp features in the output mesh. We demonstrate the capabilities of the algorithm on a variety of models and compare against state-of-the-art polyhedral meshing methods based on clipped Voronoi cells establishing the clear advantage of VoroCrust output., 18 pages (including appendix), 18 figures. Version without compressed images available on https://www.dropbox.com/s/qc6sot1gaujundy/VoroCrust.pdf. Supplemental materials available on https://www.dropbox.com/s/6p72h1e2ivw6kj3/VoroCrust_supplemental_materials.pdf

Algorithms for line and line segment clipping are well known algorithms especially in the field of computer graphics. They are formulated for the Euclidean space representation. However, computer graphics uses the projective extension of the Euclidean space and homogeneous coordinates for representation geometric transformations with points in the E² or E³ space. The projection operation from the E³ to the E² space leads to the necessity to convert coordinates to the Euclidean space if the clipping operation is to be used. In this contribution, an optimized simple algorithm for line and line segment clipping in the E² space, which works directly with homogeneous representation and not requiring the conversion to the Euclidean space, is described. It is based on Geometric Algebra (GA) formulation for projective representation. The proposed algorithm is simple, efficient and easy to implement. The algorithm can be efficiently modified for the SSE4 instruction use or the GPU application, too.

The calculation of sunlit surfaces in a building has always been a relevant aspect in building energy simulation programs. Due to the high computational cost, some programs use algorithms for shading calculation for certain solar positions after discretization of hemispherical sky. The influence of the level of discretization on the estimation of incident direct radiation on building surfaces, as well as on the required computational times, are studied in this work. The direct solar energy on a window for a year, with simulation time steps of five minutes, has been simulated by using an algorithm based on Projection and Clipping Methods. A total of 6144 simulations have been carried out, varying window sizes, window orientations, typologies of shading devices, latitudes and discretization levels of the hemispherical sky. In terms of annual incident solar energy, the results show that maximum error values are about 5% for a low level of angular discretization. Errors up to 22% in hourly incident solar energy have been estimated for some of the configurations analysed. Furthermore, a great number of configurations show errors of shading factor on a window of up to 30%, which could be most relevant in studies of natural lighting. The study also shows that the improvement achieved by the most accurate discretization level implies an increase in computational cost of about 30 times.

This paper presents an efficient gradient projection-based method for structural topological optimization problems characterized by a nonlinear objective function which is minimized over a feasible region defined by bilateral bounds and a single linear equality constraint. The specialty of the constraints type, as well as heuristic engineering experiences are exploited to improve the scaling scheme, projection, and searching step. In detail, gradient clipping and a modified projection of searching direction under certain condition are utilized to facilitate the efficiency of the proposed method. Besides, an analytical solution is proposed to approximate this projection with negligible computation and memory costs. Furthermore, the calculation of searching steps is largely simplified. Benchmark problems, including the MBB, the force inverter mechanism, and the 3D cantilever beam are used to validate the effectiveness of the method. The proposed method is implemented in MATLAB which is open-sourced for educational usage., The two authors contributed equally to this work. This paper has been accepted by Advances in Engineering Software

Multi-directional 3D printing has the capability of decreasing or eliminating the need for support structures. Recent work proposed a beam-guided search algorithm to find an optimized sequence of plane-clipping, which gives volume decomposition of a given 3D model. Different printing directions are employed in different regions to fabricate a model with tremendously less support (or even no support in many cases).To obtain optimized decomposition, a large beam width needs to be used in the search algorithm, leading to a very time-consuming computation. In this paper, we propose a learning framework that can accelerate the beam-guided search by using a smaller number of the original beam width to obtain results with similar quality. Specifically, we use the results of beam-guided search with large beam width to train a scoring function for candidate clipping planes based on six newly proposed feature metrics. With the help of these feature metrics, both the current and the sequence-dependent information are captured by the neural network to score candidates of clipping. As a result, we can achieve around 3x computational speed. We test and demonstrate our accelerated decomposition on a large dataset of models for 3D printing., Comment: 8 pages, accepted by IEEE Robotics and Automation Letters 2020

Mammography is still considered the best screening method for detection, diagnosis and follow-up of breast cancer. A correct classification of mammographic findings demands a high expertise level of the clinician observer. For this, different Computer-aided Diagnosis systems have been developed to support the diagnosis tasks and reduce the inter or intra-observer variability caused by the complex visual information contained in mammograms. However, the classification of some findings (masses, calcifications) is still a difficult task. This work presents a methodological approach to evaluate the performance of the training process for different convolutional neural network configurations of the VGG16 Convolutional Neural Network architecture, designed to perform mammographic classification. For doing that, the impact of different learning strategies (focal loss, to deal with highly unbalance datasets, gradient clipping and learning transfer) is evaluated.

Recent work on Renyi Differential Privacy has shown the feasibility of applying differential privacy to deep learning tasks. Despite their promise, however, differentially private deep networks often lag far behind their non-private counterparts in accuracy, showing the need for more research in model architectures, optimizers, etc. One of the barriers to this expanded research is the training time -- often orders of magnitude larger than training non-private networks. The reason for this slowdown is a crucial privacy-related step called "per-example gradient clipping" whose naive implementation undoes the benefits of batch training with GPUs. By analyzing the back-propagation equations we derive new methods for per-example gradient clipping that are compatible with auto-differentiation (e.g., in PyTorch and TensorFlow) and provide better GPU utilization. Our implementation in PyTorch showed significant training speed-ups (by factors of 54x - 94x for training various models with batch sizes of 128). These techniques work for a variety of architectural choices including convolutional layers, recurrent networks, attention, residual blocks, etc., Comment: To appear in PETS2021

This paper proposes a novel deep reinforcement learning architecture that was inspired by previous tree structured architectures which were only useable in discrete action spaces. Policy Prediction Network offers a way to improve sample complexity and performance on continuous control problems in exchange for extra computation at training time but at no cost in computation at rollout time. Our approach integrates a mix between model-free and model-based reinforcement learning. Policy Prediction Network is the first to introduce implicit model-based learning to Policy Gradient algorithms for continuous action space and is made possible via the empirically justified clipping scheme. Our experiments are focused on the MuJoCo environments so that they can be compared with similar work done in this area.

Deep learning models bear the risks of privacy leakage. Attackers can obtain sensitive information contained in training data with some techniques. However, existing differentially private methods such as Differential Privacy-Stochastic Gradient Descent (DP-SGD) and Differential Privacy-Generative Adversarial Network (DP-GAN) are not very efficient as they require to perform sampling multiple times. More importantly, DP-GAN algorithm need public data to set gradient clipping threshold. In this paper, we introduce our refined algorithms to tackle these problems. First, we employ random shuffling instead of random sampling to improve training efficiency. We also test Gaussian and Laplace Mechanisms for clipping gradients and injecting noise. Second, we employ zero Concentrated Differential Privacy (zCDP) to compute overall privacy budget. Finally, we adopt dynamical gradient clipping in DP-GAN algorithm. During each iteration, we random sample training examples and set the average gradients norm as the new threshold. This not only makes the algorithm more robust but also doesn’t increase the overall privacy budget. We experiment with our algorithms on MNIST data sets and demonstrate the accuracies. In our refined DP-SGD algorithm, we achieve test accuracy of 96.58%. In our refined DP-GAN algorithm, we adopt the synthetic data to train models and reach test accuracy of 91.64%. The results show that our approach ensures model usability and provides the capability of privacy protection.