摘要 :
We demonstrate three kinds of optoelectronic devices used to all-optical signal processing, namely, semiconductor optical amplifier (SOA), periodically poled lithium niobate (PPLN), and microring resonator. Especially all these de...
展开
We demonstrate three kinds of optoelectronic devices used to all-optical signal processing, namely, semiconductor optical amplifier (SOA), periodically poled lithium niobate (PPLN), and microring resonator. Especially all these devices are competitive for all-optical wavelength conversion, logic gates, and format conversion, etc.
收起
摘要 :
Reconstructing cone-beam computed tomography (CBCT) typically utilizes a Feldkamp-Davis-Kress (FDK) algorithm to 'translate' hundreds of 2D X-ray projections on different angles into a 3D CT image. For minimizing the X-ray induced...
展开
Reconstructing cone-beam computed tomography (CBCT) typically utilizes a Feldkamp-Davis-Kress (FDK) algorithm to 'translate' hundreds of 2D X-ray projections on different angles into a 3D CT image. For minimizing the X-ray induced ionizing radiation, sparse-view CBCT takes fewer projections by a wider-angle interval, but suffers from an inferior CT reconstruction quality. To solve this, the recent solutions mainly resort to synthesizing missing projections, and force the synthesized projections to be as realistic as those actual ones, which is extremely difficult due to X-ray's tissue superimposing. In this paper, we argue that the synthetic projections should restore FDK-required information as much as possible, while the visual fidelity is the secondary importance. Inspired by a simple fact that FDK only relies on frequency information after ramp-filtering for reconstruction, we develop a Reconstruction-Friendly Interpolation Network (RFI-Net), which first utilizes a 3D-2D attention network to learn inter-projection relations for synthesizing missing projections, and then introduces a novel Ramp-Filter loss to constrain a frequency consistency between the synthesized and real projections after ramp-filtering. By doing so, RFI-Net's energy can be forcibly devoted to restoring more CT-reconstruction useful information in projection synthesis. We build a complete reconstruction framework consisting of our developed RFI-Net, FDK and a commonly-used CT post-refinement. Experimental results on reconstruction from only one-eighth projections demonstrate that using RFI-Net restored full-view projections can significantly improve the reconstruction quality by increasing PSNR by 2.59 dB and 2.03 dB on the walnut and patient CBCT datasets, respectively, comparing with using those restored by other state-of-the-arts.
收起
摘要 :
Image matting aims to estimate the opacity of foreground objects in order to accurately extract them from the background. Existing methods are only concerned with RGB features to obtain alpha mattes, limiting the perception of loc...
展开
Image matting aims to estimate the opacity of foreground objects in order to accurately extract them from the background. Existing methods are only concerned with RGB features to obtain alpha mattes, limiting the perception of local tiny details. To address this issue, we introduce frequency information as an auxiliary clue to accurately distinguish foreground boundaries and propose the Frequency Matting Network (FMN). Specifically, we deploy a Frequency Boosting Module (FBM) in addition to the Discrete Cosine Transform (DCT) to extract frequency information from input images. The proposed FBM is a learn-able component that empowers the model to adapt to complex scenarios. Furthermore, we design a Domain Aggregation Module (DAM) to effectively fuse frequency features with RGB features. With the assistance of frequency clues, our proposed FMN achieves significant improvements in matting accuracy and visual quality compared with state-of-the-art methods. Extensive experiments on Composition-1k and Distinctions-646 datasets demonstrate the superiority of introducing frequency information for image matting.
收起
摘要 :
Disaggregated Persistent Memory (DPM) is a promising technology offering elasticity, high resource utilization, persistent data storage, and lower power consumption. While building KV stores on the DPM benefits from these merits, ...
展开
Disaggregated Persistent Memory (DPM) is a promising technology offering elasticity, high resource utilization, persistent data storage, and lower power consumption. While building KV stores on the DPM benefits from these merits, achieving efficient writes also faces two primary challenges: 1) limited scalability caused by the underused PM bandwidth, and 2) limited CPU resources the persistent memory server (PMS) can provide. Integrating the SmartNIC such as the Data Processing Unit (DPU) into the DPM gives developers the chance to optimize writing to KV stores by utilizing both the memory and processor of DPU. However, simple offloading cannot make full use of the DPU’s potential capacity. To address these challenges, we propose DoW-KV, a persistent hash KV store on DPM. DoW-KV employs a two-tier hash index consisting of a DPU cache table in DPU memory and multiple PM persistent tables on the PM. It relocates small random writes to the DPU memory and consolidates them to the PM at a coarse granularity. Furthermore, DoW-KV uses DPU-offloaded step merge and a coroutine-based asynchronous processing framework to efficiently manage the PM persistent tables. DoW-KV also introduces a client-mixed read strategy to boost key searching on the two-tier hash index. Experimental results show that DoW-KV outperforms the state-of-the-art DINOMO by 2.1× and 1.3× in the Put and Get operations, respectively.
收起
摘要 :
With the increasing need for analyzing graph data, graph systems have to efficiently deal with concurrent graph processing (CGP) jobs. However, existing platforms are inherently designed for a single job, they incur the high cost ...
展开
With the increasing need for analyzing graph data, graph systems have to efficiently deal with concurrent graph processing (CGP) jobs. However, existing platforms are inherently designed for a single job, they incur the high cost when CGP jobs are executed. In this work, we observed that existing systems do not allow CGP jobs to share graph structure data of each iteration, introducing redundant accesses to same graph. Moreover, all the graphs are real-world graphs with highly skewed power-law degree distributions. The gain from extending multiple external storage devices is diminishing rapidly, which needs reasonable schedulings to balance I/O pressure into each storage. Following this direction, we propose GraphScSh that handles CGP jobs efficiently on a single machine, which focuses on reducing I/O conflict and sharing graph structure data among CGP jobs. We apply a CGP balanced partition method to break graphs into multiple partitions that are stored in multiple external storage devices. Additionally, we present a CGP I/O scheduling method, so that I/O conflict can be reduced and graph data can be shared among multiple jobs. We have implemented GraphScSh in C++ and the experiment shows that GraphScSh outperforms existing out-of-core systems by up to 82%.
收起
摘要 :
With the increasing need for analyzing graph data, graph systems have to efficiently deal with concurrent graph processing (CGP) jobs. However, existing platforms are inherently designed for a single job, they incur the high cost ...
展开
With the increasing need for analyzing graph data, graph systems have to efficiently deal with concurrent graph processing (CGP) jobs. However, existing platforms are inherently designed for a single job, they incur the high cost when CGP jobs are executed. In this work, we observed that existing systems do not allow CGP jobs to share graph structure data of each iteration, introducing redundant accesses to same graph. Moreover, all the graphs are real-world graphs with highly skewed power-law degree distributions. The gain from extending multiple external storage devices is diminishing rapidly, which needs reasonable schedulings to balance I/O pressure into each storage. Following this direction, we propose GraphScSh that handles CGP jobs efficiently on a single machine, which focuses on reducing I/O conflict and sharing graph structure data among CGP jobs. We apply a CGP balanced partition method to break graphs into multiple partitions that are stored in multiple external storage devices. Additionally, we present a CGP I/O scheduling method, so that I/O conflict can be reduced and graph data can be shared among multiple jobs. We have implemented GraphScSh in C++ and the experiment shows that GraphScSh outperforms existing out-of-core systems by up to 82%.
收起
摘要 :
Spatial decompositions are often used in the statistics of location information. For security, current works split the whole domain into sub-domains recursively to generate a hierarchical private tree and add Laplace noise to each...
展开
Spatial decompositions are often used in the statistics of location information. For security, current works split the whole domain into sub-domains recursively to generate a hierarchical private tree and add Laplace noise to each node's points count, as called differentially private spatial decompositions. However Laplace distribution is symmetric about the origin, the mean of a large number of queries may cancel the Laplace noise. In private tree, the point count of intermediate nodes may be real since the summation of all its descendants may cancel the Laplace noise and reveal privacy. Moreover, existing algorithms add noises to all nodes of the private tree which leads to higher noise cost, and the maximum depth h of the tree is not intuitive for users. To address these problems, we propose a more secure algorithm which avoids canceling Laplace noise. That splits the domains depending on its real point count, and only adds indefeasible Laplace noise to leaves. The ith randomly selected leaf of one intermediate node is added noise by (β-i+1)+1+β /(β-i+1)+βLap(λ). We also replace h with a more intuitive split unit u. The experiment results show that our algorithm performs better both on synthetic and real datasets with higher security and data utility, and the noise cost is highly decreased.
收起
摘要 :
Aspect-based sentiment classification (ABSC) aims to determine the sentiment polarity toward a specific aspect. In order to finish this task, it is difficult to match a specific aspect with its opinion words since there are usuall...
展开
Aspect-based sentiment classification (ABSC) aims to determine the sentiment polarity toward a specific aspect. In order to finish this task, it is difficult to match a specific aspect with its opinion words since there are usually multiple aspects with different opinion words in a sentence. Many efforts have been made to address this problem, such as graph neural networks and attention mechanism, however come at the cost of the introduced extraneous noise, leading to mismatches of the aspect with its opinion words. In this paper, we propose a Mutual Information-based ABSC model, called MatchMaker, which introduces Mutual Information estimation to strengthen the correlations between a specific aspect and its opinion words without introducing any extraneous noise, thus significantly improving the accuracy when determining the sentiment polarity toward a specific aspect. Experimental results show that our method with Mutual Information is effective. For example, MatchMaker obtains a significant improvement of accuracy over ASGCN model by 3.1% on the RestI4.
收起
摘要 :
Aspect-based sentiment classification (ABSC) aims to determine the sentiment polarity toward a specific aspect. In order to finish this task, it is difficult to match a specific aspect with its opinion words since there are usuall...
展开
Aspect-based sentiment classification (ABSC) aims to determine the sentiment polarity toward a specific aspect. In order to finish this task, it is difficult to match a specific aspect with its opinion words since there are usually multiple aspects with different opinion words in a sentence. Many efforts have been made to address this problem, such as graph neural networks and attention mechanism, however come at the cost of the introduced extraneous noise, leading to mismatches of the aspect with its opinion words. In this paper, we propose a Mutual Information-based ABSC model, called MatchMaker, which introduces Mutual Information estimation to strengthen the correlations between a specific aspect and its opinion words without introducing any extraneous noise, thus significantly improving the accuracy when determining the sentiment polarity toward a specific aspect. Experimental results show that our method with Mutual Information is effective. For example, MatchMaker obtains a significant improvement of accuracy over ASGCN model by 3.1% on the Rest14.
收起
摘要 :
"Learned" admission policies have shown promise in improving Content Delivery Network (CDN) cache performance and lowering operational costs. Unfortunately, existing learned policies are optimized with a few fixed cache sizes whil...
展开
"Learned" admission policies have shown promise in improving Content Delivery Network (CDN) cache performance and lowering operational costs. Unfortunately, existing learned policies are optimized with a few fixed cache sizes while in reality, cache sizes often vary over time in an unpredictable manner. As a result, existing solutions cannot provide consistent benefits in production settings.We present SLAP, a learned CDN cache admission approach based on segmented object reuse time prediction. SLAP predicts an object’s reuse time range using the Long-Short-Term-Memory model and admits objects that will be reused (before eviction) given the current cache size. SLAP separates model training from cache size, allowing it to adapt to arbitrary sizes. The key to our solution is a novel segmented labeling scheme that enables SLAP to precisely predict object reuse time. To further make SLAP a practical and efficient solution, we propose aggressive reusing of computation and training on sampled traces to optimize model training, and a specialized predictor architecture that overlaps prediction computation with miss object fetching to optimize model inference. Our experiments with production CDN traces show that SLAP achieves significantly lower write traffic (38%-59%), longer SSDs service life (104%-178%), a consistently higher hit rate (3.2%-11.7%), and requires no effort to adapt to changing cache sizes, outperforming existing policies.
收起