Machine-Learning tasks are becoming pervasive in a broad range of domains, and systems (from embedded to data centers). At the same time, small set machine-learning algorithms (especially Convolutional Deep Neural Networks, i.e., CNNs DNNs) proving be state-of-the-art across many applications. As architectures evolve towards heterogeneous multi-cores composed mix cores accelerators, accelerator can achieve rare combination efficiency (due number target algorithms) application scope.

Recently, optical neural networks (ONNs) integrated into photonic chips have received extensive attention because they are expected to implement the same pattern recognition tasks in electronic platforms with high efficiency and low power consumption. However, there no efficient learning algorithms for training of ONNs on an on-chip integration system. In this article, we propose a novel strategy based neuroevolution design train ONNs. Two typical used determine hyper-parameters optimize...

In the Big Data era, gap between storage performance and an application's I/O requirement is increasing. congestion caused by concurrent accesses from multiple applications inevitable severely harms performance. Conventional approaches either focus on optimizing access pattern individually or handle requests a low-level layer without any knowledge upper-level applications. this paper, we present novel I/O-aware batch scheduling framework to coordinate ongoing petascale computing systems. The...

Gaussian Splatting (GS) methods, including 3DGS and 2DGS, have demonstrated significant effectiveness in real-time novel view synthesis (NVS), establishing themselves as a key technology the field of computer graphics. However, GS-based methods still face challenges rendering high-quality image details. Even when utilizing advanced frameworks, their outputs may display artifacts relying solely on few input views, such noise blurriness. A reasonable approach is to conduct post-processing...

Social media sites such as Twitter continue to grow at a fast pace. People of all generations use social exchange messages and share experiences their life in timely fashion. Most these make data available. An intriguing question is can we exploit this real-time massive data-flow improve business measurable way. In paper, are particularly interested tweets (Twitter messages) that relevant mobile network performance. We compare with more traditional source user experience, i.e., customer care...

Torus-connected network is widely used in modern supercomputers due to its linear per node cost scaling and competitive overall performance. Job scheduling system plays a critical role for the efficient use of supercomputers. As continue growing size, fundamental problem arises: how effectively balance job performance with on torus-connected machines? In this work, we will present new design named window-based locality-aware scheduling. Our contains three novel features. First, rather than...

As HPC systems scale toward exascale, it becomes critical to manage the underlying resource more effectively. While almost all existing management schedule jobs in a queuing fashion and have drawbacks of making isolated scheduling decisions that would compromise system performance even with backfilling, plan-based schedulers potential generate better job schedules by producing an execution plan waiting but do not receive enough attention. In this paper, we present novel utilizes simulated...

Although home automation (HA) systems in the wired domain are widely accepted by consumers, today's industry, mega trend is steering HA along a wireless way. Theoretically, solutions able to provide with more flexibility and thus reducing engineering costs. In practice, however, deploying actually requires costs efforts due lack of versatile software tools support whole process. This paper defines evaluates workflow architecture for systems. The proposed studied implemented based on web...

Federated learning enables a collaborative training and optimization of global models among group devices without sharing local data samples. However, the heterogeneity in federated can lead to unfair representation model across different devices. To address fairness issue learning, we propose dynamic q algorithm with reinforcement called DQFFL. DQFFL aims mitigate discrepancies device aggregation enhance treatment for all groups involved learning. quantify fairness, leverages performance on...

A new terrain meshing method based on the ray-tracing algorithm is presented. In propagation model, most of transmission paths are focused in area that between transmitters and receivers. The main idea to mesh containing both receivers into fine triangles while rest larger triangles. processing time this developed half traditional maintaining accuracy. great number simulations an outdoor environment were conducted get these results. significance theory practice.

In this article, we propose a novel method using machine learning, especially for artificial neural networks (ANNs) to achieve variability analysis and performance optimization of the plasmonic refractive index sensor (RIS). A Fano resonance (FR) based RIS which consisted two waveguides end-coupled each other by an asymmetrical square resonator is taken as illustration demonstrate effectiveness ANNs. The results reveal that ANNs can be used in fast accurate because predicted transmission...

In spite of great efforts to expand the network infrastructure, it is still hard catch rapid growing need serve tremendous subscribers with significant traffic. To effectively allocate resource among mobile telecommunication operators and Internet Service Providers (ISP), grasping highly dynamic traffic patterns can enable effective planning optimization for state-of-the-art Internet. We employ behavior analysis method analyze model Due subscribers' number streaming media accounting over 60%...

The effect of the graded In_xGa_1-xAs layer on distribution strain was studied by calculating different models using finite element method. results demonstrate that can reduce and thus lead to longer emission wavelength. also increase in GaAs capping which cause disadvantage grow stacked InAs/GaAs QD structure. But be released though thickness spacer when

중국 현대 도시 조형물은 증가 추세에 있으며, 문화의 중요한 부분이 되었다. 그러나 도시조형물의 조형성에 대한 전문적인 연구가 부족하고, 감상자들의 디자인 평가 요인이 부재하다. 이에 본 연구는 조형물에 이미지 요인을 도출하고자 하였다. 빅데이터 분석을 통해 어휘를 추출한 후, 설문조사의 통계분석을 요인 및 선호도를 분석하였다. 분석 결과 첫째, 조형물의 어휘 43개를 도출하였으며, 설문조사 SPSS통계분석을 혁신성, 조형성, 주제성, 도시성, 인문성, 관리성 등 6가지 추출하였다. 둘째, 인구 특성에 따른 결과, 학력에서는 모두 유의하며, 직업에서는 주제성과 그리고 관리성에서 유의하였고, 연령에서는 도시성에서 유의하였다. 셋째, 내적 어휘의 차이 주제에서는

During the process of wrapper scan chain balance, it is inevitable that a certain number idle bits will appear for inequality in length between different chains. In this paper, new balance algorithm called best exchange optimization (BEO) presented first. This aims at minimizing Later test scheduling method bit percentage on bus (IBPTB)-based (ITS) proposed. ITS generates optimal solution according to IBPTBs rectangles each IP core. The benchmark circuits ITC'02 are selected be objects...

The paper presents a reverse SoC TAM design based dual-balanced strategy, which is on the basis of IEEE1500. Firstly test scheduling executed according to conceptual architecture that physically realizable, and then real can be reversely established result. Since not limited by architecture, optimization involve both top level IP obtain cross-level combined between these two levels. experimental results ITCpsila02 show better availability reliability performance improvement time proposed...