arXiv subjects

Zheng Cao

Publications and source records attributed to Zheng Cao.

51 records · Page 3Linked to original sources

arXiv2021

AIBench Scenario: Scenario-distilling AI Benchmarking

Modern real-world application scenarios like Internet services consist of a diversity of AI and non-AI modules with huge code sizes and long and complicated execution paths, which raises serious benchmarking or evaluating challenges. Using AI components or micro benchmarks alone can lead to error-prone conclusions. This paper presents a methodology to attack the above challenge. We formalize a real-world application scenario as a Directed Acyclic Graph-based model and propose the rules to distill it into a permutation of essential AI and non-AI tasks, which we call a scenario benchmark. Together with seventeen industry partners, we extract nine typical scenario benchmarks. We design and implement an extensible, configurable, and flexible benchmark framework. We implement two Internet service AI scenario benchmarks based on the framework as proxies to two real-world application scenarios. We consider scenario, component, and micro benchmarks as three indispensable parts for evaluating. Our evaluation shows the advantage of our methodology against using component or micro AI benchmarks alone. The specifications, source code, testbed, and results are publicly available from \url{https://www.benchcouncil.org/aibench/scenario/}.

cs.PF↗

arXiv2021

AIBench Training: Balanced Industry-Standard AI Training Benchmarking

Earlier-stage evaluations of a new AI architecture/system need affordable benchmarks. Only using a few AI component benchmarks like MLPerfalone in the other stages may lead to misleading conclusions. Moreover, the learning dynamics are not well understood, and the benchmarks' shelf-life is short. This paper proposes a balanced benchmarking methodology. We use real-world benchmarks to cover the factors space that impacts the learning dynamics to the most considerable extent. After performing an exhaustive survey on Internet service AI domains, we identify and implement nineteen representative AI tasks with state-of-the-art models. For repeatable performance ranking (RPR subset) and workload characterization (WC subset), we keep two subsets to a minimum for affordability. We contribute by far the most comprehensive AI training benchmark suite. The evaluations show: (1) AIBench Training (v1.1) outperforms MLPerfTraining (v0.7) in terms of diversity and representativeness of model complexity, computational cost, convergent rate, computation, and memory access patterns, and hotspot functions; (2) Against the AIBench full benchmarks, its RPR subset shortens the benchmarking cost by 64%, while maintaining the primary workload characteristics; (3) The performance ranking shows the single-purpose AI accelerator like TPU with the optimized TensorFlowframework performs better than that of GPUs while losing the latter's general support for various AI models. The specification, source code, and performance numbers are available from the AIBench homepage https://www.benchcouncil.org/aibench-training/index.html.

cs.AI↗

arXiv2020

Lightweight Convolutional Neural Networks for CSI Feedback in Massive MIMO

In frequency division duplex mode of massive multiple-input multiple-output systems, the downlink channel state information (CSI) must be sent to the base station (BS) through a feedback link. However, transmitting CSI to the BS is costly due to the bandwidth limitation of the feedback link. Deep learning (DL) has recently achieved remarkable success in CSI feedback. Realizing high-performance and low-complexity CSI feedback is a challenge in DL based communication. We develop a DL based CSI feedback network in this study to complete the feedback of CSI effectively. However, this network cannot be effectively applied to the mobile terminal because of the excessive numbers of parameters. Therefore, we further propose a new lightweight CSI feedback network based on the developed network. Simulation results show that the proposed CSI network exhibits better reconstruction performance than that of other CsiNet-related works. Moreover, the lightweight network maintains a few parameters and parameter complexity while ensuring satisfactory reconstruction performance. These findings suggest the feasibility and potential of the proposed techniques.

cs.IT↗

arXiv2020

AIBench: An Agile Domain-specific Benchmarking Methodology and an AI Benchmark Suite

Domain-specific software and hardware co-design is encouraging as it is much easier to achieve efficiency for fewer tasks. Agile domain-specific benchmarking speeds up the process as it provides not only relevant design inputs but also relevant metrics, and tools. Unfortunately, modern workloads like Big data, AI, and Internet services dwarf the traditional one in terms of code size, deployment scale, and execution path, and hence raise serious benchmarking challenges. This paper proposes an agile domain-specific benchmarking methodology. Together with seventeen industry partners, we identify ten important end-to-end application scenarios, among which sixteen representative AI tasks are distilled as the AI component benchmarks. We propose the permutations of essential AI and non-AI component benchmarks as end-to-end benchmarks. An end-to-end benchmark is a distillation of the essential attributes of an industry-scale application. We design and implement a highly extensible, configurable, and flexible benchmark framework, on the basis of which, we propose the guideline for building end-to-end benchmarks, and present the first end-to-end Internet service AI benchmark. The preliminary evaluation shows the value of our benchmark suite---AIBench against MLPerf and TailBench for hardware and software designers, micro-architectural researchers, and code developers. The specifications, source code, testbed, and results are publicly available from the web site \url{http://www.benchcouncil.org/AIBench/index.html}.

cs.PF↗

arXiv2019

Algorithmic Design and Implementation of Unobtrusive Multistatic Serial LiDAR Image

To fully understand interactions between marine hydrokinetic (MHK) equipment and marine animals, a fast and effective monitoring system is required to capture relevant information whenever underwater animals appear. A new automated underwater imaging system composed of LiDAR (Light Detection and Ranging) imaging hardware and a scene understanding software module named Unobtrusive Multistatic Serial LiDAR Imager (UMSLI) to supervise the presence of animals near turbines. UMSLI integrates the front end LiDAR hardware and a series of software modules to achieve image preprocessing, detection, tracking, segmentation and classification in a hierarchical manner.

eess.IV↗

arXiv2019

AIBench: An Industry Standard Internet Service AI Benchmark Suite

Today's Internet Services are undergoing fundamental changes and shifting to an intelligent computing era where AI is widely employed to augment services. In this context, many innovative AI algorithms, systems, and architectures are proposed, and thus the importance of benchmarking and evaluating them rises. However, modern Internet services adopt a microservice-based architecture and consist of various modules. The diversity of these modules and complexity of execution paths, the massive scale and complex hierarchy of datacenter infrastructure, the confidential issues of data sets and workloads pose great challenges to benchmarking. In this paper, we present the first industry-standard Internet service AI benchmark suite---AIBench with seventeen industry partners, including several top Internet service providers. AIBench provides a highly extensible, configurable, and flexible benchmark framework that contains loosely coupled modules. We identify sixteen prominent AI problem domains like learning to rank, each of which forms an AI component benchmark, from three most important Internet service domains: search engine, social network, and e-commerce, which is by far the most comprehensive AI benchmarking effort. On the basis of the AIBench framework, abstracting the real-world data sets and workloads from one of the top e-commerce providers, we design and implement the first end-to-end Internet service AI benchmark, which contains the primary modules in the critical paths of an industry scale application and is scalable to deploy on different cluster scales. The specifications, source code, and performance numbers are publicly available from the benchmark council web site http://www.benchcouncil.org/AIBench/index.html.

cs.CV↗

arXiv2018

The impact of $S$-wave thresholds $D_{s1}\bar{D}_{s}+c.c.$ and $D_{s0}\bar{D}^*_{s}+c.c.$ on vector charmonium spectrum

By investigating the very closely lied $D_{s1}\bar{D}_{s}+c.c.$ and $D_{s0}\bar{D}^*_{s}+c.c.$ thresholds at about 4.43 GeV we propose that the $ψ(4415)$ and $ψ(4160)$ can be mixing states between the dynamic generated states of the strong $S$-wave $D_{s1}\bar{D}_{s}+c.c.$ and $D_{s0}\bar{D}^*_{s}+c.c.$ interactions and the quark model states $ψ(4S)$ and $ψ(2D)$. We investigate the $J/ψK\bar{K}$ final states and invariant mass spectrum of $J/ψK$ to demonstrate that nontrivial lineshapes can arise from such a mechanism. This process, which goes through triangle loop transitions, is located in the vicinity of the so-called "triangle singularity (TS)" kinematics. As a result, it provides a special mechanism for the production of exotic states $Z_{cs}$, which is the strange partner of $Z_c(3900)$, but with flavor contents of $c\bar{c}q\bar{s}$ (or $c\bar{c}s\bar{q}$) with $q$ denoting $u/d$ quarks. The lineshapes of the $e^+e^-\to J/ψK\bar{K}$ cross sections and $J/ψK \ (J/ψ\bar{K})$ spectrum are sensitive to the dynamically generated state, and we demonstrate that a pole structure can be easily distinguished from open threshold CUSP effects if an exotic state is created. A precise measurement of the cross section lineshapes can test such a mixing mechanism and provide navel information for the exotic partners of the $Z_c(3900)$ in the charmonium spectrum.

hep-ph↗

arXiv2018

BigDataBench: A Scalable and Unified Big Data and AI Benchmark Suite

Several fundamental changes in technology indicate domain-specific hardware and software co-design is the only path left. In this context, architecture, system, data management, and machine learning communities pay greater attention to innovative big data and AI algorithms, architecture, and systems. Unfortunately, complexity, diversity, frequently-changed workloads, and rapid evolution of big data and AI systems raise great challenges. First, the traditional benchmarking methodology that creates a new benchmark or proxy for every possible workload is not scalable, or even impossible for Big Data and AI benchmarking. Second, it is prohibitively expensive to tailor the architecture to characteristics of one or more application or even a domain of applications. We consider each big data and AI workload as a pipeline of one or more classes of units of computation performed on different initial or intermediate data inputs, each class of which we call a data motif. On the basis of our previous work that identifies eight data motifs taking up most of the run time of a wide variety of big data and AI workloads, we propose a scalable benchmarking methodology that uses the combination of one or more data motifs---to represent diversity of big data and AI workloads. Following this methodology, we present a unified big data and AI benchmark suite---BigDataBench 4.0, publicly available from~\url{http://prof.ict.ac.cn/BigDataBench}. This unified benchmark suite sheds new light on domain-specific hardware and software co-design: tailoring the system and architecture to characteristics of the unified eight data motifs other than one or more application case by case. Also, for the first time, we comprehensively characterize the CPU pipeline efficiency using the benchmarks of seven workload types in BigDataBench 4.0.

cs.DC↗

arXiv2018

Anomaly Analysis for Co-located Datacenter Workloads in the Alibaba Cluster

In warehouse-scale cloud datacenters, co-locating online services and offline batch jobs is an efficient approach to improving datacenter utilization. To better facilitate the understanding of interactions among the co-located workloads and their real-world operational demands, Alibaba recently released a cluster usage and co-located workload dataset, which is the first publicly dataset with precise information about the category of each job. In this paper, we perform a deep analysis on the released Alibaba workload dataset, from the perspective of anomaly analysis and diagnosis. Through data preprocessing, node similarity analysis based on Dynamic Time Warping (DTW), co-located workloads characteristics analysis and anomaly analysis based on iForest, we reveals several insights including: (1) The performance discrepancy of machines in Alibaba's production cluster is relatively large, for the distribution and resource utilization of co-located workloads is not balanced. For instance, the resource utilization (especially memory utilization) of batch jobs is fluctuating and not as stable as that of online containers, and the reason is that online containers are long-running jobs with more memory-demanding and most batch jobs are short jobs, (2) Based on the distribution of co-located workload instance numbers, the machines can be classified into 8 workload distribution categories1. And most patterns of machine resource utilization curves are similar in the same workload distribution category. (3) In addition to the system failures, unreasonable scheduling and workload imbalance are the main causes of anomalies in Alibaba's cluster.

cs.DC↗

arXiv2018

RELXILL_NK: a relativistic reflection model for testing Einstein's gravity

Einstein's theory of general relativity was proposed over 100 years ago and has successfully passed a large number of observational tests in the weak field regime. However, the strong field regime is largely unexplored, and there are many modified and alternative theories that have the same predictions as Einstein's gravity for weak fields and present deviations when gravity becomes strong. RELXILL_NK is the first relativistic reflection model for probing the spacetime metric in the vicinity of astrophysical black holes and testing Einstein's gravity in the strong field regime. Here we present our current constraints on possible deviations from Einstein's gravity obtained from the black holes in 1H0707-495, Ark 564, GX 339-4, and GS 1354-645.

gr-qc↗

arXiv2018

Testing general relativity with the reflection spectrum of the supermassive black hole in 1H0707$-$495

Recently, we have extended the X-ray reflection model relxill to test the spacetime metric in the strong gravitational field of astrophysical black holes. In the present Letter, we employ this extended model to analyze XMM-Newton, NuSTAR, and Swift data of the supermassive black hole in 1H0707-495 and test deviations from a Kerr metric parametrized by the Johannsen deformation parameter $α_{13}$. Our results are consistent with the hypothesis that the spacetime metric around the black hole in 1H0707-495 is described by the Kerr solution.

gr-qc↗

arXiv2017

Marine Animal Classification with Correntropy Loss Based Multi-view Learning

To analyze marine animals behavior, seasonal distribution and abundance, digital imagery can be acquired by visual or Lidar camera. Depending on the quantity and properties of acquired imagery, the animals are characterized as either features (shape, color, texture, etc.), or dissimilarity matrices derived from different shape analysis methods (shape context, internal distance shape context, etc.). For both cases, multi-view learning is critical in integrating more than one set of feature or dissimilarity matrix for higher classification accuracy. This paper adopts correntropy loss as cost function in multi-view learning, which has favorable statistical properties for rejecting noise. For the case of features, the correntropy loss-based multi-view learning and its entrywise variation are developed based on the multi-view intact space learning algorithm. For the case of dissimilarity matrices, the robust Euclidean embedding algorithm is extended to its multi-view form with the correntropy loss function. Results from simulated data and real-world marine animal imagery show that the proposed algorithms can effectively enhance classification rate, as well as suppress noise under different noise conditions.

cs.CV↗

arXiv2017

Testing conformal gravity with astrophysical black holes

Weyl conformal symmetry can solve the problem the spacetime singularities present in Einstein's gravity. In a recent paper, two of us have found a singularity-free rotating black hole solution in conformal gravity. In addition to the mass $M$ and the spin angular momentum $J$ of the black hole, the new solution has a new parameter, $L$, which here we consider to be proportional to the black hole mass. Since the solution is conformally equivalent to the Kerr metric, photon trajectories are unchanged, while the structure of an accretion disk around a black hole is affected by the value of the parameter $L$. In this paper, we show that X-ray data of astrophysical black holes require $L/M < 1.2$.

gr-qc↗

arXiv2016

Open charm contributions to the E1 transitions of $ψ(3686)$ and $ψ(3770)\to γχ_{cJ}$

The E1 transitions of $ψ(3686)$ and $ψ(3770)\to γχ_{cJ}$ are investigated in a non-relativistic effective field theory (NREFT) where the open charm effects are included systematically as the leading corrections. It also allows a self-consistent inclusion of the $S$-$D$ mixing in the same framework. We are able to show that the open charm contributions are essential for understanding the significant discrepancies between the non-relativistic leading order calculations and the experimental data.

hep-ph↗

arXiv2016

Iron K$α$ line of boson stars

The present paper is a sequel to our previous work [Y. Ni et al., JCAP 1607, 049 (2016)] in which we studied the iron K$α$ line expected in the reflection spectrum of Kerr black holes with scalar hair. These metrics are solutions of Einstein's gravity minimally coupled to a massive, complex scalar field. They form a continuous bridge between a subset of Kerr black holes and a family of rotating boson stars depending on one extra parameter, the dimensionless scalar hair parameter $q$, ranging from 0 (Kerr black holes) to 1 (boson stars). Here we study the limiting case $q=1$, corresponding to rotating boson stars. For comparison, spherical boson stars are also considered. We simulate observations with XIS/Suzaku. Using the fact that current observations are well fit by the Kerr solution and thus requiring that acceptable alternative compact objects must be compatible with a Kerr fit, we find that some boson star solutions are relatively easy to rule out as potential candidates to explain astrophysical black holes, while other solutions, which are neither too dilute nor too compact are more elusive and we argue that they cannot be distinguished from Kerr black holes by the analysis of the iron line with current X-ray facilities.

gr-qc↗