arXiv subjects

Dongsheng Wang

Publications and source records attributed to Dongsheng Wang.

61 records · Page 4Linked to original sources

arXiv2019

QGAN: Quantized Generative Adversarial Networks

The intensive computation and memory requirements of generative adversarial neural networks (GANs) hinder its real-world deployment on edge devices such as smartphones. Despite the success in model reduction of CNNs, neural network quantization methods have not yet been studied on GANs, which are mainly faced with the issues of both the effectiveness of quantization algorithms and the instability of training GAN models. In this paper, we start with an extensive study on applying existing successful methods to quantize GANs. Our observation reveals that none of them generates samples with reasonable quality because of the underrepresentation of quantized values in model weights, and the generator and discriminator networks show different sensitivities upon quantization methods. Motivated by these observations, we develop a novel quantization method for GANs based on EM algorithms, named as QGAN. We also propose a multi-precision algorithm to help find the optimal number of bits of quantized GAN models in conjunction with corresponding result qualities. Experiments on CIFAR-10 and CelebA show that QGAN can quantize GANs to even 1-bit or 2-bit representations with results of quality comparable to original models.

cs.NE↗

arXiv2017

Computation Error Analysis of Block Floating Point Arithmetic Oriented Convolution Neural Network Accelerator Design

The heavy burdens of computation and off-chip traffic impede deploying the large scale convolution neural network on embedded platforms. As CNN is attributed to the strong endurance to computation errors, employing block floating point (BFP) arithmetics in CNN accelerators could save the hardware cost and data traffics efficiently, while maintaining the classification accuracy. In this paper, we verify the effects of word width definitions in BFP to the CNN performance without retraining. Several typical CNN models, including VGG16, ResNet-18, ResNet-50 and GoogLeNet, were tested in this paper. Experiments revealed that 8-bit mantissa, including sign bit, in BFP representation merely induced less than 0.3% accuracy loss. In addition, we investigate the computational errors in theory and develop the noise-to-signal ratio (NSR) upper bound, which provides the promising guidance for BFP based CNN engine design.

cs.LG↗

arXiv2016

Symmetry-protected topological phases with uniform computational power in one dimension

We investigate the usefulness of ground states of quantum spin chains with symmetry-protected topological order (SPTO) for measurement-based quantum computation. We show that, in spatial dimension one, if an SPTO phase supports quantum wire, then, subject to an additional symmetry condition that is satisfied in all cases so far investigated, it can also be used for quantum computation.

quant-ph↗

arXiv2016

MAC: a novel systematically multilevel cache replacement policy for PCM memory

The rapid development of multi-core system and increase of data-intensive application in recent years call for larger main memory. Traditional DRAM memory can increase its capacity by reducing the feature size of storage cell. Now further scaling of DRAM faces great challenge, and the frequent refresh operations of DRAM can bring a lot of energy consumption. As an emerging technology, Phase Change Memory (PCM) is promising to be used as main memory. It draws wide attention due to the advantages of low power consumption, high density and nonvolatility, while it incurs finite endurance and relatively long write latency. To handle the problem of write, optimizing the cache replacement policy to protect dirty cache block is an efficient way. In this paper, we construct a systematically multilevel structure, and based on it propose a novel cache replacement policy called MAC. MAC can effectively reduce write traffic to PCM memory with low hardware overhead. We conduct simulation experiments on GEM5 to evaluate the performances of MAC and other related works. The results show that MAC performs best in reducing the amount of writes (averagely 25.12%) without increasing the program execution time.

cs.AR↗

arXiv2015

Uncertainty Principle Respects Locality

The notion of nonlocality implicitly implies there might be some kind of spooky action at a distance in nature, however, the validity of quantum mechanics has been well tested up to now. In this work it is argued that the notion of nonlocality is physically improper, the basic principle of locality in nature is well respected by quantum mechanics, namely, the uncertainty principle. We show that the quantum bound on the Clauser, Horne, Shimony, and Holt (CHSH) inequality can be recovered from the uncertainty relation in a multipartite setting. We further argue that the super-quantum correlation demonstrated by the nonlocal box is not physically comparable with the quantum one. The origin of the quantum structure of nature still remains to be explained, some post-quantum theory which is more complete in some sense than quantum mechanics is possible and might not necessarily be a hidden variable theory.

physics.gen-ph↗

arXiv2013

Complete positivity and contextuality of quantum dynamics

Positivity or the stronger notion of complete positivity, and contextuality are central properties of quantum dynamics. In this work, we demonstrate that a physical unitary-universe dilation model could be employed to characterize the completely positive map, regardless of the initial correlation condition. Particularly, the problem of initial correlation can be resolved by a swap operation. Furthermore, we discuss the physical essence of completely positive map and highlights its limitations. Then we develop the quantum measurement-chain formula beyond the framework of completely positive map in order to describe much broader quantum dynamics, and therein the property of contextuality could be captured via measurement transfer matrix.

quant-ph↗

arXiv2010

Quantum interference in a four-level system of a $^{87}\mathrm{Rb}$ atom: Effects of spontaneously generated coherence

In this work, the effects of quantum interference and spontaneously generated coherence (SGC) are theoretically analyzed in a four level system of a $^{87}\mathrm{Rb}$ atom. For the effects of SGC, we find that a new kind of EIT channel can be induced due to destructive interference, and the nonlinear Kerr absorption can be coherently narrowed or eliminated under different strengths of the coupling and switching fields.

quant-ph↗