arXiv subjects

Hao Jin

Publications and source records attributed to Hao Jin.

50 records · Page 3Linked to original sources

arXiv2021

Natural Language Adversarial Defense through Synonym Encoding

In the area of natural language processing, deep learning models are recently known to be vulnerable to various types of adversarial perturbations, but relatively few works are done on the defense side. Especially, there exists few effective defense method against the successful synonym substitution based attacks that preserve the syntactic structure and semantic information of the original text while fooling the deep learning models. We contribute in this direction and propose a novel adversarial defense method called Synonym Encoding Method (SEM). Specifically, SEM inserts an encoder before the input layer of the target model to map each cluster of synonyms to a unique encoding and trains the model to eliminate possible adversarial perturbations without modifying the network architecture or adding extra data. Extensive experiments demonstrate that SEM can effectively defend the current synonym substitution based attacks and block the transferability of adversarial examples. SEM is also easy and efficient to scale to large models and big datasets.

cs.CL↗

arXiv2021

Meta-Regularization: An Approach to Adaptive Choice of the Learning Rate in Gradient Descent

We propose \textit{Meta-Regularization}, a novel approach for the adaptive choice of the learning rate in first-order gradient descent methods. Our approach modifies the objective function by adding a regularization term on the learning rate, and casts the joint updating process of parameters and learning rates into a maxmin problem. Given any regularization term, our approach facilitates the generation of practical algorithms. When \textit{Meta-Regularization} takes the $φ$-divergence as a regularizer, the resulting algorithms exhibit comparable theoretical convergence performance with other first-order gradient-based algorithms. Furthermore, we theoretically prove that some well-designed regularizers can improve the convergence performance under the strong-convexity condition of the objective function. Numerical experiments on benchmark problems demonstrate the effectiveness of algorithms derived from some common $φ$-divergence in full batch as well as online learning settings.

cs.LG↗

arXiv2021

Positive Seebeck coefficient in highly doped La$_{2-x}$Sr$_x$CuO$_4$ ($x$=0.33); its origin and implication

We present a study of the thermoelectric (Seebeck and Nernst) response in heavily overdoped, non-superconducting La$_{1.67}$Sr$_{0.33}$CuO$_4$. In spite of the electron-like curvature of the Fermi surface, the Seebeck coefficient is positive at low temperatures. Such a feature, previously observed in copper, silver, gold and lithium, is caused by a non-trivial energy dependence of the scattering time. We argue that this feature implies a strong asymmetry between the lifetime of occupied and unoccupied states along the zone diagonals and such an electron-hole asymmetry impedes formation of Cooper pairs along the nodal direction in the superconducting ground state emerging at lower doping levels.

cond-mat.supr-con↗

arXiv2020

Locating arrays with mixed alphabet sizes

Locating arrays (LAs) can be used to detect and identify interaction faults among factors in a component-based system. The optimality and constructions of LAs with a single fault have been investigated extensively under the assumption that all the factors have the same values. However, in real life, different factors in a system have different numbers of possible values. Thus, it is necessary for LAs to satisfy such requirements. We herein establish a general lower bound on the size of mixed-level $(\bar{1},t)$-locating arrays. Some methods for constructing LAs including direct and recursive constructions are provided. In particular, constructions that produce optimal LAs satisfying the lower bound are described. Additionally, some series of optimal LAs satisfying the lower bound are presented.

math.CO↗

arXiv2019

Towards Better Generalization: BP-SVRG in Training Deep Neural Networks

Stochastic variance-reduced gradient (SVRG) is a classical optimization method. Although it is theoretically proved to have better convergence performance than stochastic gradient descent (SGD), the generalization performance of SVRG remains open. In this paper we investigate the effects of some training techniques, mini-batching and learning rate decay, on the generalization performance of SVRG, and verify the generalization performance of Batch-SVRG (B-SVRG). In terms of the relationship between optimization and generalization, we believe that the average norm of gradients on each training sample as well as the norm of average gradient indicate how flat the landscape is and how well the model generalizes. Based on empirical observations of such metrics, we perform a sign switch on B-SVRG and derive a practical algorithm, BatchPlus-SVRG (BP-SVRG), which is numerically shown to enjoy better generalization performance than B-SVRG, even SGD in some scenarios of deep neural networks.

stat.ML↗

arXiv2019

Constrained locating arrays for combinatorial interaction testing

This paper introduces the notion of Constrained Locating Arrays (CLAs), mathematical objects which can be used for fault localization in software testing. CLAs extend ordinary locating arrays to make them applicable to testing of systems that have constraints on test parameters. Such constraints are common in real-world systems; thus CLA enhances the applicability of locating arrays to practical testing problems. The paper also proposes an algorithm for constructing CLAs. Experimental results show that the proposed algorithm scales to problems of practical sizes.

cs.SE↗

arXiv2019

Practical Verifiable In-network Filtering for DDoS defense

In light of ever-increasing scale and sophistication of modern DDoS attacks, it is time to revisit in-network filtering or the idea of empowering DDoS victims to install in-network traffic filters in the upstream transit networks. Recent proposals show that filtering DDoS traffic at a handful of large transit networks can handle volumetric DDoS attacks effectively. However, the innetwork filtering primitive can also be misused. Transit networks can use the in-network filtering service as an excuse for any arbitrary packet drops made for their own benefit. For example, transit networks may intentionally execute filtering services poorly or unfairly to discriminate their competing neighbor ASes while claiming that they drop packets for the sake of DDoS defense. We argue that it is due to the lack of verifiable filtering - i.e., no one can check if a transit network executes the filter rules correctly as requested by the DDoS victims. To make in-network filtering a more robust defense primitive, in this paper, we propose a verifiable in-network filtering, called VIF, that exploits emerging hardware-based trusted execution environments (TEEs) and offers filtering verifiability to DDoS victims and neighbor ASes. Our proof of concept demonstrates that a VIF filter implementation on commodity servers with TEE support can handle traffic at line rate (e.g., 10 Gb/s) and execute up to 3,000 filter rules. We show that VIF can easily scale to handle larger traffic volume (e.g., 500 Gb/s) and more complex filtering operations (e.g., 150,000 filter rules) by parallelizing the TEE-based filters. As a practical deployment model, we suggest that Internet exchange points (IXPs) are the ideal candidates for the early adopters of our verifiable filters due to their central locations and flexible software-defined architecture.

cs.CR↗

arXiv2018

Fine-Grained Endpoint-Driven In-Network Traffic Control for Proactive DDoS Attack Mitigation

Volumetric attacks, which overwhelm the bandwidth of a destination, are among the most common DDoS attacks today. Despite considerable effort made by both research and industry, our recent interviews with over 100 potential DDoS victims in over 10 industry segments indicate that today's DDoS prevention is far from perfect. On one hand, few academical proposals have ever been deployed in the Internet; on the other hand, solutions offered by existing DDoS prevention vendors are not a silver bullet to defend against the entire attack spectrum. Guided by such large-scale study of today's DDoS defense, in this paper, we present MiddlePolice, the first readily deployable and proactive DDoS prevention mechanism. We carefully architect MiddlePolice such that it requires no changes from both the Internet core and the network stack of clients, yielding instant deployability in the current Internet architecture. Further, relying on our novel capability feedback mechanism, MiddlePolice is able to enforce destination-driven traffic control so that it guarantees to deliver victim-desired traffic regardless of the attacker strategies. We implement a prototype of MiddlePolice, and demonstrate its feasibility via extensive evaluations in the Internet, hardware testbed and large-scale simulations.

cs.NI↗

arXiv2018

Observation of extremely long exciton lifetime in Janus-MoSTe monolayer

The electron-hole separation efficiency is a key factor that determines the performance of two-dimensional (2D) transition metal dichalcogenides (TMDs) and devices. Therefore, searching for novel 2D TMD materials with long timescale of carrier lifetime becomes one of the most important topics. Here, based on the timedomain density functional theory (TD-DFT), we propose a brand new TMD material, namely janus-MoSTe, which exhibits a strong build-in electric field. Our results show that in janus-MoSTe monolayer, the exciton consisting of electron and hole has a relatively wide spatial extension and low binding energy. In addition, a slow electron-hole recombination process is observed, with timescale on the order of 1.31 ns, which is even comparable with those of van der Waals (vdW) heterostructures. Further analysis reveals that the extremely long timescale for electron-hole recombination could be ascribed to the strong Coulomb screening effect as well as the small overlap of wavefunctions between electrons and holes. Our findings establish the build-in electric field as an effective factor to control the electron-hole recombination dynamics in TMD monolayers and facilitate their future applications in light detecting and harvesting.

cond-mat.mtrl-sci↗

arXiv2017

First-principles study on the electronic and transport properties of periodically nitrogen-doped graphene and carbon nanotube superlattices

Prompted by recent reports on $\sqrt{3} \times \sqrt{3}$ graphene superlattices with intrinsic inter-valley interactions, we perform first-principles calculations to investigate the electronic properties of periodically nitrogen-doped graphene and carbon nanotube nanostructures. In these structures, nitrogen atoms substitute one-sixth of the carbon atoms in the pristine hexagonal lattices with exact periodicity to form perfect $\sqrt{3} \times \sqrt{3}$ superlattices of graphene and carbon nanotubes. Multiple nanostructures of $\sqrt{3} \times \sqrt{3}$ graphene ribbons and carbon nanotubes are explored, and all configurations show nonmagnetic and metallic behaviors. The transport properties of $\sqrt{3} \times \sqrt{3}$ graphene and carbon nanotube superlattices are calculated utilizing the non-equilibrium Green's function formalism combined with density functional theory. The transmission spectrum through the pristine and $\sqrt{3} \times \sqrt{3}$ armchair carbon nanotube heterostructure shows quantized behavior under certain circumstances.

cond-mat.mes-hall↗

arXiv2004

Inherent Relation between Nernst Signal and Nodal Quasiparticle Transport in Pseudogap Region of Underdoped High Temperature Superconductors

In-plane Nernst signal and resistivity have been measured for three $La_{2-x}Sr_xCuO_4$ single crystals (x=0.09, 0.11 and 0.145) with the magnetic field parallel to c-axis. A quadratic temperature dependence of resistivity, i.e., $ρ=ρ_0+aT^2$ is observed below a certain temperature $T_R$. It is found that the upper boundary of the Nernst signal $T_n$ coincides with $T_R$, which points to an inherent relation between the anomalous Nernst signal and the nodal quasiparticle transport in the pseudogap region. Finally a phase diagram together with the pseudogap temperature $T^*$ is presented, which suggests a second energy scale in the pseudogap region.

cond-mat.supr-con↗

arXiv2004

Thermoelectricity of URu$_{2}$Si$_{2}$: giant Nernst effect in the hidden-order state

We present a study of Nernst and Seebeck coefficients in the heavy-fermion compound URu$_{2}$Si$_{2}$, which hosts a phase transition of unsettled origin. A giant Nernst signal of unprecedented magnitude was found to emerge in the ordered state. Moreover, our analysis of the Seebeck and Hall data indicates that the ordering leads to a sudden increase in the entropy per itinerant electron and to a drastic decrease in the scattering rate.

cond-mat.str-el↗

arXiv2003

Rigid vortices in MgB2

Magnetic relaxation of high-pressure synthesized MgB$_2$ bulks with different thickness is investigated. It is found that the superconducting dia-magnetic moment depends on time in a logarithmic way; the flux-creep activation energy decreases linearly with the current density (as expected by Kim-Anderson model); and the activation energy increases linearly with the thickness of sample when it is thinner than about 1 mm. These features suggest that the vortices in the MgB$_2$ are rather rigid, and the pinning and creep can be well described by Kim-Anderson model.

cond-mat.supr-con↗

arXiv2003

Vortex-unbinding and finite-size effects in Tl2Ba2CaCu2O8 thin films

Current-voltage ($I$-$V$) characteristics of Tl$_2$Ba$_2$CaCu$_2$O$_8$ thin films in zero magnetic field are measured and analyzed with the conventional Kosterlitz-Thouless-Berezinskii (KTB) approach, dynamic scaling approach and finite-size scaling approach, respectively. It is found from these results that the $I$-$V$ relation is determined by the vortex-unbinding mechanism with the KTB dynamic critical exponent $z=2$. On the other hand, the evidence of finite-size effect is also found, which blurs the feature of a phase transition.

cond-mat.supr-con↗