arXiv subjects
Publications and source records attributed to Pengcheng Li.
In this article, we consider the finite exceptional groups of Lie type $E_6$ and $^2E_6$. We prove the inductive blockwise Alperin weight condition holds for unipotent $l$-blocks of $E_6^{\varepsilon}(q)$ if $2,3\nmid q$, $l\geq 5$
Production software oftentimes suffers from the issue of performance inefficiencies caused by inappropriate use of data structures, programming abstractions, and conservative compiler optimizations. It is desirable to avoid unnecessary memory operations. However, existing works often use a whole-program fine-grained monitoring method with incredibly high overhead. To this end, we propose a learning-aided approach to identify unnecessary memory operations intelligently with low overhead. By applying several prevalent graph neural network models to extract program semantics with respect to program structure, execution order and dynamic states, we present a novel, hybrid program embedding approach so that to derive unnecessary memory operations through the embedding. We train our model with tens of thousands of samples acquired from a set of real-world benchmarks. Results show that our model achieves 90% of accuracy and incurs only around a half of time overhead of the state-of-art tool.
With data durability, high access speed, low power efficiency and byte addressability, NVMe and SSD, which are acknowledged representatives of emerging storage technologies, have been applied broadly in many areas. However, one key issue with high-performance adoption of these technologies is how to properly define intelligent cache layers such that the performance gap between emerging technologies and main memory can be well bridged. To this end, we propose Phoebe, a reuse-aware reinforcement learning framework for the optimal online caching that is applicable for a wide range of emerging storage models. By continuous interacting with the cache environment and the data stream, Phoebe is capable to extract critical temporal data dependency and relative positional information from a single trace, becoming ever smarter over time. To reduce training overhead during online learning, we utilize periodical training to amortize costs. Phoebe is evaluated on a set of Microsoft cloud storage workloads. Experiment results show that Phoebe is able to close the gap of cache miss rate from LRU and a state-of-the-art online learning based cache policy to the Belady's optimal policy by 70.3% and 52.6%, respectively.
The impact parameter is one of the crucial physical quantities of heavy-ion collisions (HICs), and can affect obviously many observables at the final state, such as the multifragmentation and the collective flow. Usually, it cannot be measured directly in experiments but might be inferred from observables at the final state. Artificial intelligence has had great success in learning complex representations of data, which enables novel modeling and data processing approaches in physical sciences. In this article, we employ two of commonly used algorithms in the field of artificial intelligence, the Convolutional Neural Networks (CNN) and Light Gradient Boosting Machine (LightGBM), to improve the accuracy of determining impact parameter by analyzing the proton spectra in transverse momentum and rapidity on the event-by-event basis. Au+Au collisions with the impact parameter of 0$\leq$$b$$\leq$10 fm at intermediate energies ($E_{\rm lab}$=$0.2$-$1.0$ GeV$/$nucleon) are simulated with the ultrarelativistic quantum molecular dynamics (UrQMD) model to generate the proton spectra data. It is found that the average difference between the true impact parameter and the estimated one can be smaller than 0.1 fm. The LightGBM algorithm shows an improved performance with respect to the CNN on the task in this work. By using the LightGBM's visualization algorithm, one can obtain the important feature map of the distribution of transverse momentum and rapidity, which may be helpful in inferring the impact parameter or centrality in heavy-ion experiments.
TianQin is a planned space-based gravitational wave (GW) observatory consisting of three earth orbiting satellites with an orbital radius of about $10^5~{\rm km}$. The satellites will form a equilateral triangle constellation the plane of which is nearly perpendicular to the ecliptic plane. TianQin aims to detect GWs between $10^{-4}~{\rm Hz}$ and $1~{\rm Hz}$ that can be generated by a wide variety of important astrophysical and cosmological sources, including the inspiral of Galactic ultra-compact binaries, the inspiral of stellar-mass black hole binaries, extreme mass ratio inspirals, the merger of massive black hole binaries, and possibly the energetic processes in the very early universe or exotic sources such as cosmic strings. In order to start science operations around 2035, a roadmap called the 0123 plan is being used to bring the key technologies of TianQin to maturity, supported by the construction of a series of research facilities on the ground. Two major projects of the 0123 plan are being carried out. In this process, the team has created a new generation $17~{\rm cm}$ single-body hollow corner-cube retro-reflector which has been launched with the QueQiao satellite on 21 May 2018; a new laser ranging station equipped with a $1.2~{\rm m}$ telescope has been constructed and the station has successfully ranged to all the five retro-reflectors on the Moon; and the TianQin-1 experimental satellite has been launched on 20 December 2019 and the first round result shows that the satellite has exceeded all of its mission requirements.
The (co)homology self-closeness number of a simply-connected based CW-complexes $X$ is the minimal number $k$ such that any self-map $f$ of $X$ inducing an automorphism of the (co)homology groups for dimensions$\leq k$ is a self-homotopy equivalence. These two numbers are homotopy invariants and have a close relation with the group of self-homotopy equivalences. In this paper, we compare the (co)homology self-closeness numbers of spaces in certain cofibrations, define the mod $p$ (co)homology self-closeness number of simply-connected $p$-local spaces with finitely generated homologies and study some properties of the (mod $p$) (co)homology self-closeness numbers.
Caching techniques are widely used in the era of cloud computing from applications, such as Web caches to infrastructures, Memcached and memory caches in computer architectures. Prediction of cached data can greatly help improve cache management and performance. The recent advancement of deep learning techniques enables the design of novel intelligent cache replacement policies. In this work, we propose a learning-aided approach to predict future data accesses. We find that a powerful LSTM-based recurrent neural network model can provide high prediction accuracy based on only a cache trace as input. The high accuracy results from a carefully crafted locality-driven feature design. Inspired by the high prediction accuracy, we propose a pseudo OPT policy and evaluate it upon 13 real-world storage workloads from Microsoft Research. Results demonstrate that the new cache policy improves state-of-art practical policies by up to 19.2% and incurs only 2.3% higher miss ratio than OPT on average.
The state-of-the-art deep learning algorithms rely on distributed training systems to tackle the increasing sizes of models and training data sets. Minibatch stochastic gradient descent (SGD) algorithm requires workers to halt forward/back propagations, to wait for gradients aggregated from all workers, and to receive weight updates before the next batch of tasks. This synchronous execution model exposes the overheads of gradient/weight communication among a large number of workers in a distributed training system. We propose a new SGD algorithm, DaSGD (Local SGD with Delayed Averaging), which parallelizes SGD and forward/back propagations to hide 100% of the communication overhead. By adjusting the gradient update scheme, this algorithm uses hardware resources more efficiently and reduces the reliance on the low-latency and high-throughput inter-connects. The theoretical analysis and the experimental results show its convergence rate O(1/sqrt(K)), the same as SGD. The performance evaluation demonstrates it enables a linear performance scale-up with the cluster size.
The difference in elliptic flow $v_{2}$ between protons and antiprotons, produced in $^{197}\text{Au}+^{197}\text{Au}$ collisions at center-of-mass energies $\sqrt{s_{NN}}=5-12~\text{GeV}$, is studied within a modified version of the ultrarelativistic quantum molecular dynamics (UrQMD) model. Two different model scenarios are compared: the cascade mode and the mean field mode which includes potential interactions for both formed and pre-formed hadrons. The model results for the elliptic flow of protons and the relative $v_{2}$ difference between protons and antiprotons obtained from the mean field mode agree with the available experimental data, while the $v_{2}$ difference is near zero for the cascade mode. Our results show that the elliptic flow splitting, observed for particles and antiparticles, can be explained by the inclusion of proper hadronic interactions. In addition, the difference in $v_{2}$ between protons and antiprotons depends on the centrality and the rapidity window. With smaller centrality and/or rapidity acceptance, the observed elliptic flow splitting is more sensitive to the beam energy, indicating a strong net baryon density dependence of the effect. We propose to confirm this splitting at the upcoming experiments from Beam Energy Scan (BES) Phase-\Rmnum{2} at Relativistic Heavy Ion Collider (RHIC), the Compressed Baryonic Matter (CBM) at Facility for Antiproton and Ion Research (FAIR), High Intensity heavy ion Accelerator Facility (HIAF) and Nuclotron-based Ion Collider fAcility (NICA).
Recent studies have highlighted that deep neural networks (DNNs) are vulnerable to adversarial examples. In this paper, we improve the robustness of DNNs by utilizing techniques of Distance Metric Learning. Specifically, we incorporate Triplet Loss, one of the most popular Distance Metric Learning methods, into the framework of adversarial training. Our proposed algorithm, Adversarial Training with Triplet Loss (AT$^2$L), substitutes the adversarial example against the current model for the anchor of triplet loss to effectively smooth the classification boundary. Furthermore, we propose an ensemble version of AT$^2$L, which aggregates different attack methods and model structures for better defense effects. Our empirical studies verify that the proposed approach can significantly improve the robustness of DNNs without sacrificing accuracy. Finally, we demonstrate that our specially designed triplet loss can also be used as a regularization term to enhance other defense methods.
Deep neural network (DNN) as a popular machine learning model is found to be vulnerable to adversarial attack. This attack constructs adversarial examples by adding small perturbations to the raw input, while appearing unmodified to human eyes but will be misclassified by a well-trained classifier. In this paper, we focus on the black-box attack setting where attackers have almost no access to the underlying models. To conduct black-box attack, a popular approach aims to train a substitute model based on the information queried from the target DNN. The substitute model can then be attacked using existing white-box attack approaches, and the generated adversarial examples will be used to attack the target DNN. Despite its encouraging results, this approach suffers from poor query efficiency, i.e., attackers usually needs to query a huge amount of times to collect enough information for training an accurate substitute model. To this end, we first utilize state-of-the-art white-box attack methods to generate samples for querying, and then introduce an active learning strategy to significantly reduce the number of queries needed. Besides, we also propose a diversity criterion to avoid the sampling bias. Our extensive experimental results on MNIST and CIFAR-10 show that the proposed method can reduce more than $90\%$ of queries while preserve attacking success rates and obtain an accurate substitute model which is more than $85\%$ similar with the target oracle.
Fluctuations of the baryon number in relativistic heavy-ion collisions are a promising observable to explore the structure of the QCD phase diagram. The cumulant ratios in heavy ion collisions at intermediate energies ($\sqrt{s_{\textrm{NN}}} < 7$ GeV) have not been studied to date. We investigate the effects of mean field potential and clustering on the cumulant ratios of baryon and proton number distributions in Au+Au collisions at beam energy of 1.23 GeV$/$nucleon as measured by the HADES Collaboration at GSI. Ultrarelativistic Quantum Molecular Dynamics (UrQMD) and the JAM model are used to calculate the cumulants with different mean field potentials. It is found that the cumulant ratios are strongly time dependent. At the early stage, the effects of the potentials on the fluctuations of the particle multiplicity in momentum space are relatively weak. The mean fields enhance the fluctuations during the expansion stage, especially for small rapidity acceptance windows. The enhancement of cumulant ratios for free protons is strongly suppressed as compared to that for all baryons. The mean field potentials and the clustering play an important role for the measured cumulant ratios at intermediate energy.
With the newly updated version of the ultrarelativistic quantum molecular dynamics (UrQMD) model, a systematic investigation of the effects of in-medium nucleon-nucleon ($NN$) elastic cross section on the collective flow and the stopping observables in $^{197}\text{Au}+^{197}\text{Au}$ collisions at beam energies from 40 to 150 MeV/nucleon is performed. Simulations with the medium correction factor $\mathcal{F}=σ^{\text{in-medium}}_{NN}/σ^{\text{free}}_{NN}=0.2,~0.3,~0.5$, and the one obtained with the FU3FP1 parametrization which depends on both the density and the momentum are compared to the FOPI and INDRA experimental data. It is found that, to best fit the experimental data of the slope of the directed flow and the elliptic flow at mid-rapidity as well as the nuclear stopping, the correction factor $\mathcal{F}$=0.2 and 0.5 are required for reactions at beam energies of 40 and 150 MeV/nucleon, respectively. While calculations with the FU3FP1 parametrization can simultaneously reproduce these experimental data reasonably well. And, the observed increasing nuclear stopping with increasing beam energy in experimental data can also be reproduced by using the FU3FP1 parametrization, while the calculated stopping power in Au+Au collisions with beam energies from 40 to 150 MeV$/$nucleon almost keeps constant when take $\mathcal{F}$ equal to a fixed value.
In this paper, we perform a cosmological model-independent test of the cosmic distance-duality relation (CDDR) in terms of the ratio of angular diameter distance (ADD) $D=D_{\rm A}^{\rm sl}/D_{\rm A}^{\,\rm s}$ from strong gravitational lensing (SGL) and the ratio of luminosity distance (LD) $D^\ast=D_{\rm L}^{\,\rm l}/D_{\rm L}^{\,\rm s}$ obtained from the joint of type Ia supernovae (SNIa) Union2.1 compilation and the latest Gamma-Ray Bursts (GRBs) data, where the superscripts s and l correspond to the redshifts $z_{\,\rm s}$ and $z_{\,\rm l}$ at the source and lens from SGL samples. The purpose of combining GRB data with SNIa compilation is to test CDDR in a wider redshift range. The LD associated with the redshits of the observed ADD, is obtained through two cosmological model-independent methods, namely, method A: binning the SNIa+GRBs data, and method B: reconstructing the function of DL by combining the Crossing Statistic with the smoothing method. We find that CDDR is compatible with the observations at $1σ$ confidence level for the power law model which is assumed to describe the mass distribution of lensing systems with method B in a wider redshift range.
Einstein's General Relativity theory simplifies dramatically in the limit that the spacetime dimension D is very large. This could still be true in the gravity theory with higher derivative terms. In this paper, as the first step to study the gravity with a Gauss-Bonnet(GB) term, we compute the quasi-normal modes of the spherically symmetric GB black hole in the large D limit. When the GB parameter is small, we find that the non-decoupling modes are the same as the Schwarzschild case and the decoupled modes are slightly modified by the GB term. However, when the GB parameter is large, we find some novel features. We notice that there are another set of non-decoupling modes due to the appearance of a new plateau in the effective radial potential. Moreover, the effective radial potential for the decoupled vector-type and scalar-type modes becomes more complicated. Nevertheless we manage to compute the frequencies of the these decoupled modes analytically. When the GB parameter is neither very large nor very small, though analytic computation is not possible, the problem is much simplified in the large D expansion and could be numerically treated. We study numerically the vector-type quasinormal modes in this case.
I review the work of proximity effect correction (PEC) in electron-beam (e-beam) lithography with emphasis on dose modification and shape modification PEC techniques.
Electron-beam (e-beam) lithography is widely employed in fabrication of 2-D patterns and 3-D structures. A certain type or shape of the sidewall in the remaining resist profile may be desired in an application, e.g., an undercut for lift-off and a vertical sidewall for etching, or required for a device. Also, as the feature size is decreased well below a micron, a small variation of the sidewall slope can lead to a significant (relative) CD error in certain layers of resist. Therefore, it is important to understand effects of spatial dose distribution on sidewall shape and be able to achieve the desired shape. In this study, via simulation, the relationship among the total dose, spatial distribution of dose, developing time and sidewall shape, and performance of the method developed to optimize the dose distribution for a target sidewall shape have been analyzed. The simulation results have been verified through experiments.
We report a systematic study of normal state Nernst effect in the electron-doped cuprates Pr$_{2-x}$Ce$_x$CuO$_{4-δ}$ over a wide range of doping (0.05$\leq x \leq$0.21) and temperature. At low temperatures, we observed a notable vortex Nernst signal above T$_c$ in the underdoped films, but no such normal state vortex Nernst signal is found in the overdoped region. The superconducting fluctuations in the underdoped region are most likely incoherent phase fluctuations as found in hole-doped cuprates. At high temperatures, a large normal state Nernst signal is found at dopings from slightly underdoped to highly overdoped. Combined with normal state thermoelectric power, Hall effect and magnetoresistance measurements, the large Nernst effect is compatible with two-band model. For the highly overdoped films, the large Nernst effect is anomalous and not explainable with a simple hole-like Fermi surface seen in photoemission experiments.