Search

In healthcare, it is essential to explain the decision-making process of machine learning models to establish the trustworthiness of clinicians. This paper introduces BI-RADS-Net, a novel explainable deep learning approach for cancer detection in breast ultrasound images. The proposed approach incorporates tasks for explaining and classifying breast tumors, by learning feature representations relevant to clinical diagnosis. Explanations of the predictions (benign or malignant) are provided in terms of morphological features that are used by clinicians for diagnosis and reporting in medical practice. The employed features include the BI-RADS descriptors of shape, orientation, margin, echo pattern, and posterior features. Additionally, our approach predicts the likelihood of malignancy of the findings, which relates to the BI-RADS assessment category reported by clinicians. Experimental validation on a dataset consisting of 1,192 images indicates improved model accuracy, supported by explanations in clinical terms using the BI-RADS lexicon.

Breast cancer investigation is of great significance, and developing tumor detection methodologies is a critical need. However, it is challenging for breast cancer detection using breast ultrasound (BUS) images due to the complicated breast structure and poor quality of the images. This paper proposes a novel tumor saliency estimation (TSE) model guided by enriched breast anatomy knowledge to localize the tumor. First, the breast anatomy layers are generated by a deep neural network. Then we refine the layers by integrating a non-semantic breast anatomy model to solve the problems of incomplete mammary layers. Meanwhile, a new background map generation method weighted by the semantic probability and spatial distance is proposed to improve the performance. The experiment demonstrates that the proposed method with the new background map outperforms four state-of-the-art TSE models with an increasing 10% of $F_{measure}$ on the public BUS dataset.

Deep learning approaches have achieved impressive results in breast ultrasound (BUS) image segmentation. However, these methods did not solve uncertainty and noise in BUS images well. Meanwhile, they did not involve the context information of BUS images, either. To address this issue, we present a novel deep learning structure for BUS image semantic segmentation by analyzing the uncertainty using a pyramid fuzzy block and generating a novel feature based on connectedness. There are three major contributions in this paper: (1) the structure of pyramid fuzzy block; (2) a novel membership function based on multi-convolution layers; and (3) a novel context feature based on connectedness. The proposed methods are applied to two datasets: a BUS image benchmark with two categories (background and tumor) and a five-category BUS image dataset with fat layer, mammary layer, muscle layer, background, and tumor. The proposed method achieves the best results on both datasets compared with eight state-of-the-art deep learning-based approaches.

The Orbital Angular Momentum (OAM) is one of the intrinsic properties carried by vortex electromagnetic wave, and it can be applied to future communication systems. In this paper, a single OAM communication link model is built, and corresponding performances are analyzed and simulated. Simulation results shows that the OAM can be used for actual communications.

Lowered power loss and asymmetrical reference voltage are reported in the DC degradation of ZnO varistors in this paper. Based on the frequency domain dielectric responses of the pristine and degraded samples, the present study explores the roles of point defects in the degradation process via dielectric relaxations and their activation energies. It is found that the degradation leads to the decrease of the activation energies for the two relaxations under high temperature. Given the lowest migration barrier for Zn i (0.57 eV) and high conduction of oxygen ion in Bi-rich phase, it is speculated that Zn i and O ad ″ migrate under DC bias, and then change the defect structure and the double Schottky barrier (DSB) at grain boundaries, during which the reverse-biased barrier height gradually increases, leading to the lowering of power loss.

To deploy a CNN on resource-constrained edge platforms, channel pruning techniques promise a significant reduction of implementation costs including memory, computation, and energy consumption without special hardware or software libraries. This paper proposes CHaPR, a novel pruning technique to structurally prune the redundant channels in a trained deep Convolutional Neural Network. CHaPR utilizes a proposed subset selection problem formulation for pruning which it solves using pivoted QR factorization. CHaPR also includes an additional pruning technique for ResNet-like architectures which resolves the issue encountered by some existing channel pruning methods that not all the layers can be pruned. Experimental results on VGG-16 and ResNet-50 models show 4.29X and 2.84X reduction, respectively in computation cost while incurring 2.50% top-1 and 1.40% top-5 accuracy losses. Compared to many existing works, CHaPR performs better when considering an Overall Score metric which accounts for both computation and accuracy.

Due to the multi-functional characteristic of the 5G, the design of 5G devices become more and more complicated. To this end, we designed three sets of test experiments to study the transmission performance of 5G mobile phones, and added a hand model to simulate the impact of human hands on the mobile phone. We tested and analyzed 5G terminals in the N78 band under the SA network. The measurement result is evaluated by the effective isotropic radiated power (EIRP). The measurement results show that under the SA network, the left and right hands will have different impacts on the performance of 5G mobile phones, but the impact is very small and the overall performance is stable.

Different infrastructure resources are utilized by the Cloud broker to meet the demands of the users so that the monetary cost is minimized. To a certain extent, the demands of the users can be collected or forecasted in advance, and some off-line algorithms can be used to plan the schemes to use the resources. But it is unavoidable some demands will be unpredictable and thus the input information will be inaccurate. This blunts greatly the practical effectiveness the off-line algorithms. It is impracticable to apply the off-line algorithm with great fluctuation caused by the sudden demands. This paper addresses the challenge by an online algorithm. Extensive real world traces driven evaluations show that the proposed algorithm can save cost up to 14%.

This paper investigates network representation learning which involves network structure and nodes' labels. Most methods proposed so far try to utilize different kinds of network data available in just one perfect model to learn the set of perfect embeddings, and then evaluate its performance against other methods via network analysis tasks, such as node classification. But in this paper, we do not follow this pattern. Inspired by the idea of ensemble learning, we subtly design two separate neural models to learn from network structure and nodes' labels respectively. Then we integrate embeddings of our two models and find that when the precondition of ensemble learning is satisfied, set of integrated embeddings can perform better than any of its components. Experimental results show that our method works very well and the integrated embeddings generally achieve better performance for node classification than many other excellent baselines under almost all our experimental settings.

Debugging is an essential part of the development process, and the functionality of the debugging tools influences the quality of software development. Therefore, a suitable debugging tool should be adopted before the formal development process. Distinct from the traditional software debugging, debugging android kernel requires specific tools. However, the commonly used android kernel debugging tools cannot meet the requirement of tracing the error of kernel codes during the early stage of the kernel boot process. Even if some methods are adjusted for the situation, the unneglectable cost is considered as the blocking of kernel development. In this paper, a feasible android kernel debugging method is proposed and implemented. In this method, the debugging information is stored in the non-volatile memory, which can be acquired after the kernel boot procedure. As a result, the data in this memory space will not be cleared and will not be arbitrarily modified by the kernel, and developers can trace the kernel code by analyzing the information dumped from the memory space. The evaluation of the scheme proves that this method is feasible and functional for android kernel debugging during the kernel boot process.

Analyzing large volumes of data is becoming more and more important in various scientific computing domains. MapReduce over MPI frameworks are an appealing solution to enable scalable big data analytics on supercomputing systems. These systems can further leverage features of MapReduce applications by merging (key/value) pairs before the reduce function in combiner optimizations. In this paper, we propose a pipeline combiner workflow and integrate it into Mimir, a cutting-edge implementation of Map Reduce over MPI. Our results with real datasets on the Tianhe-2 supercomputer prove that our pipeline combiner workflow can reduce memory usage up to 51 % and improve the overall performance up to 61 %.

Automatic tumor segmentation of breast ultrasound (BUS) image is quite challenging due to the complicated anatomic structure of breast and poor image quality. Most tumor segmentation approaches achieve good performance on BUS images collected in controlled settings; however, the performance degrades greatly with BUS images from different sources. Tumor saliency estimation (TSE) has attracted increasing attention to solving the problem by modeling radiologists' attention mechanism. In this paper, we propose a novel hybrid framework for TSE, which integrates both high-level domain-knowledge and robust low-level saliency assumptions and can overcome drawbacks caused by direct mapping in traditional TSE approaches. The new framework integrated the Neutro-Connectedness (NC) map, the adaptive-center, the correlation and the layer structure-based weighted map. The experimental results demonstrate that the proposed approach outperforms state-of-the-art TSE methods., Comment: 6 pages and 8 figures, Conference paper and accepted by ICPR 2018

Computer-aided diagnosis (CAD) can help doctors in diagnosing breast cancer. Breast ultrasound (BUS) imaging is harmless, effective, portable, and is the most popular modality for breast cancer detection/diagnosis. Many researchers work on improving the performance of CAD systems. However, there are two main shortcomings: (1) Most of the existing methods are based on prerequisites that there is one and only one tumor in the image. (2) The results depend on the datasets, i.e., an algorithm using different datasets may obtain different performances. It implies that the performance of traditional methods is dataset-dependent. In this paper, we propose an effective approach: (1) using information extended images to train a fully convolutional network (FCN) to semantically segment BUS image into 3 categories: mammary layer, tumor, and background; and (2) applying layer structure information - the breast cancers are located inside the mammary layer - to the conditional random field (CRF) for conducting breast cancer segmentation and making the segmentation result more accurate. The proposed method is evaluated utilizing BUS images of 325 cases, and the result is the best comparing with that of the existing methods by achieving true positive rate 92.80%, false positive rate 9%, and Intersection over Union 82.11%. The proposed approach has solved the above mentioned two shortcomings of the existing methods.

This paper presents a novel automatic indoor three-dimensional (3D) scene reconstruction approach which is guided by a beforehand two-dimensional (2D) grid map. The proposed system collects only a few images with a RGB-D sensor (Kinect v2) at poses precalculated by the grid map to reconstruct a complete model of the indoor environment. To remove noises result from mirror reflection, a wall detection based cloud filtering method is proposed as preprocessing for the point clouds. Then, robot location information provided by a laser scanner is utilized as an initial guess of the point cloud registration which uses Iterative Closest Point (ICP). Finally, to maintain global consistency and improve the overall accuracy, a graph optimization strategy is applied. Experimental results are provided to show the effectiveness of the proposed approach.

Perpendicular magnetic tunnel junctions (MTJs) based on MgO/CoFeB structures are of particular interest for spin-transfer torque magnetic random access memories (STT-MRAMs). However, their major challenges of combining both a large tunnel magneto-resistance ratio (TMR) and a low junction resistance, a strong interfacial perpendicular magnetic anisotropy (PMA) for the high thermal stability and a low STT switching critical current density are still to be met. In this paper, we show our recent progress in studying the pertinence of capping layers to several principal spintronic effects such as PMA, TMR and Gilbert damping. And we predict and experimentally prove the performance optimization the by the interfacial property tuning of heavy metal/CoFeB.

In this paper we present Mimir, a new implementation of MapReduce over MPI. Mimir inherits the core principles of existing MapReduce frameworks, such as MR-MPI, while redesigning the execution model to incorporate a number of sophisticated optimization techniques that achieve similar or better performance with significant reduction in the amount of memory used. Consequently, Mimir allows significantly larger problems to be executed in memory, achieving large performance gains. We evaluate Mimir with three benchmarks on two highend platforms to demonstrate its superiority compared with that of other frameworks.

This work is the first to propose a technology mapping algorithm for All Spin Logic (ASL) device. The ASL is the most actively-pursed one among spintronic devices which themselves fall under emerging post-CMOS nano-technologies. We identify the shortcomings of directly applying classical technology mapping with ASL devices, and propose techniques to extend it to handle these shortcomings. Our results show that our ASL-aware technology mapping algorithm can achieve on-average 9.15% and up to 27.27% improvement in delay (when optimizing delay) with slight improvement in area, compared to the solution generated by classical technology mapping. In a broader sense, our results show the need for developing circuit-level CAD tools that are aware of and optimized for emerging nano-technologies in order to better assess their promise as we move to the post-CMOS era.

Research data infrastructure such as storage must now accommodate new requirements resulting from trends in research data management that require researchers to store their data for the long term and make it available to other researchers. We propose Data Depot, a system and service that provides capabilities for shared space within a group, shared applications, flexible access patterns and ease of transfer at Purdue University. We evaluate Depot as a solution for storing and sharing multi-terabytes of data produced in the long tail of science with a use case in soundscape ecology studies from the Human-Environment Modeling and Analysis Laboratory. We observe that with the capabilities enabled by Data Depot, researchers can easily deploy fine-grained data access control, manage data transfer and sharing, as well as integrate their workflows into a High Performance Computing environment.

This paper investigates the coverage optimization of a visual sensor network for monitoring three-dimensional (3-D) environment or objects. Different from existing works, we improve the system performance from two aspects: (1) a parallel visual occlusion detection algorithm is implemented with Graphic Processing Units (GPUs) in order to increase the computing efficiency and a further improved parallel visual occlusion detection algorithm is introduced to reduce the complexity of the problem; (2) Differential Evolution (DE) is first applied to optimize deployment configurations of visual sensor networks. Comparative evaluation results demonstrate the superior performance of the proposed approach.

Controlling the magnetism by means of electric fields is a key issue for the future development of low-power spintronics1. Usually, electric-field can control the magnetic materials by oxide-gate structures2. But this approach has a low regulatory capacity drawback than a way of controlled on BaTiO 3 (BTO) by the magnetoelectricity coupling, such as the experiments of FeRh/BaTiO 3 3, BaTiO 3 /La 0.67 Sr 0.33 MnO4 and Pb(Mg 1/3 Nb 2/3 ) 0.7 Ti 0.3 O 3 /Co 40 Fe 40 B 20 5. FeGa alloy and BTO oxide are laminated to form heterostructure for high magnetoelectricity6. Ga could contribute to preferential magnetization along the direction in FeGa alloys and thus create anisotropy in the material, which is higher than the pure Fe7. Here we report the magnetic control of FeGa thin films on BaTiO substrate by electric field.

We present a scalable method to extensively search for and accurately select pharmaceutical drug candidates in large spaces of drug conformations computationally generated and stored across the nodes of a large distributed system. For each ligand conformation in the dataset, our method first extracts relevant geometrical properties and transforms the properties into a single metadata point in the three-dimensional space. Then, it performs an octree-based clustering on the metadata to search for predominant clusters. Our method avoids the need to move ligand conformations among nodes because it extracts relevant data properties locally and concurrently. By doing so, we can perform accurate and scalable distributed clustering analysis on large distributed datasets. We scale the analysis of our pharmaceutical datasets a factor of 400X higher in performance and 500X larger in size than ever before. We also show that our clustering achieves higher accuracy compared with that of traditional clustering methods and conformational scoring based on minimum energy.

The emergence of Big Data applications provides new challenges in data management such as processing and movement of masses of data. Volunteer computing has proven itself as a distributed paradigm that can fully support Big Data generation. This paradigm uses a large number of heterogeneous and unreliable Internet-connected hosts to provide Peta-scale computing power for scientific projects. With the increase in data size and number of devices that can potentially join a volunteer computing project, the host bandwidth can become a main hindrance to the analysis of the data generated by these projects, especially if the analysis is a concurrent approach based on either in-situ or in-transit processing. In this paper, we propose a bandwidth model for volunteer computing projects based on the real trace data taken from the Docking@Home project with more than 280,000 hosts over a 5-year period. We validate the proposed statistical model using model-based and simulation-based techniques. Our modeling provides us with valuable insights on the concurrent integration of data generation with in-situ and in-transit analysis in the volunteer computing paradigm.

Magnetic memory stores data by magnetic state. It provides non-volatility, fast write/read speed and infinite endurance. Currently, the data-writing approach for magnetic memory is based on the spin transfer torque effect. However, a relatively large current is needed to change the magnetic state, which would cause high power dissipation and poor scaling performance. The solution to relieve these problems is a new data-writing approach based on voltage control. A new structure of this approach with higher storage density and better energy efficiency for the magnetic memory is studied, which makes it a strong competitor among the non-volatile memories.

This paper presents a one-pass, distributed method that enables in-situ data analysis for large protein folding trajectory datasets by executing sufficiently fast, avoiding moving trajectory data, and limiting the memory usage. First, the method extracts the geometric shape features of each protein conformation in parallel. Then, it classifies sets of consecutive conformations into meta-stable and transition stages using a probabilistic hierarchical clustering method. Lastly, it rebuilds the global knowledge necessary for the intraand inter-trajectory analysis through a reduction operation. The comparison of our method with a traditional approach for a villin headpiece sub domain shows that our method generates significant improvements in execution time, memory usage, and data movement. Specifically, to analyze the same trajectory consisting of 20,000 protein conformations, our method runs in 41.5 seconds while the traditional approach takes approximately 3 hours, uses 6.9MB memory per core while the traditional method uses 16GB on one single node where the analysis is performed, and communicates only 4.4KB while the traditional method moves the entire dataset of 539MB. The overall results in this paper support our claim that our method is suitable for in-situ data analysis of folding trajectories.

In this paper, we present two variations of a general analysis algorithm for large datasets residing in distributed memory systems. Both variations avoid the need to move data among nodes because they extract relevant data properties locally and concurrently and transform the analysis problem (e.g., clustering or classification) into a search for property aggregates. We test the two variations using the SDSC's supercomputer Gordon, the MapReduce-MPI library, and a structural biology dataset of 100 million protein-ligand records. We evaluate both variations for their sensitivity to data distribution and load imbalance. Our observations indicate that the first variation is sensitive to data content and distribution while the second variation is not. Moreover, the second variation can self-heal load imbalance and it outperforms the first in all the fifteen cases considered.

Focusing on the video text detection, which is challenging and with wide potential applications, a novel stroke width feature is proposed and a system which detects text regions based on multi-scale corner detection is implemented in this paper. In our system, candidate text regions are generated by applying morphologic operation based on corner points detected in different scales, and non-text regions are filtered by combining proposed stroke width feature with some simple geometric properties. Moreover, there is a new multi-instance semi-supervised learning strategy being proposed in this paper considering the unknown contrast parameter in stroke width extraction. Experiments taken on video frames from different kinds of video shots prove that the proposed approach is both efficient and accurate for video text detection.

Ribonucleic acid (RNA) molecules play important roles in many biological processes including gene expression and regulation. Their secondary structures are crucial for the RNA functionality, and the prediction of the secondary structures is widely studied. Previous research shows that cutting long sequences into shorter chunks, predicting secondary structures of the chunks independently using thermodynamic methods, and reconstructing the entire secondary structure from the predicted chunk structures tend to yield better accuracy than predicting the secondary structure using the entire RNA sequence as a whole. The chunking, prediction, and reconstruction processes can use different methods and parameters, some of which produce more accurate predictions than others. The RNA sequence can be cut into chunks using different cutting methods and chunk lengths. Several prediction methods, with different degree of accuracy and computing requirements, can be used. The reconstruction of shorter predictions into the entire sequence can rely on simply gluing the parts together or on using more sophisticated merging algorithms. To allow scientists to perform a systematic analysis of the impact of the several methods and parameters, we propose a modularized framework using MapReduce. The framework enables scientists to automatically and efficiently explore large parametric spaces of chunking, prediction, reconstruction, and analysis methods. This paper shows how the MapReduce framework allows scientists to gain insights about different chunking strategies easily, accurately, and efficiently.

We propose a linear clustering approach for large datasets of molecular geometries produced by high-throughput molecular dynamics simulations (e.g., protein folding and protein-ligand docking simulations). To this scope, we transform each three-dimensional (3D) molecular conformation into a single point in the 3D space reducing the space complexity while still encoding the molecular similarities and geometries. We assign an identifier to each single 3D point mapping a docked ligand, generate a tree from the whole space, and apply a tree-based clustering on the reduced conformation space that identifies most dense hyperspaces. We adapt our method for MapReduce and implement it in Hadoop. The load-balancing, fault-tolerance, and scalability in MapReduce allows screening of very large conformation datasets not approachable with traditional clustering methods. We analyze results for datasets with different concentrations of optimal solutions, and draw conclusions about the limitations and usability of our method. The advantages of this approach make it attractive for complex applications in real-world high-throughput molecular simulations.