Searcharxiv⌕ Search

arXiv subjects

Jie Zhou

Publications and source records attributed to Jie Zhou.

At least 865 records · Page 48Linked to original sources
arXiv2019

A Gromov-Hausdorff convergence theorem of surfaces in $\mathbb{R}^n$ with small total curvature

In this paper, we mainly study the compactness and local structure of immersing surfaces in $\mathbb{R}^n$ with local uniform bounded area and small total curvature $\int_{Σ\cap B_1(0)} |A|^2$. A key ingredient is a new quantity which we call isothermal radius. Using the estimate of the isothermal radius we establish a compactness theorem of such surfaces in intrinsic $L^p$-topology and extrinsic $W^{2,2}$-weak topology. As applications, we can explain Leon Simon's decomposition theorem\cite{LS} in the viewpoint of convergence and prove a non-collapsing version of Hélein's convergence theorem\cite{H}\cite{KL12}.

math.DG↗
arXiv2019

Restricting Control Flow During Speculative Execution with Venkman

Side-channel attacks such as Spectre that utilize speculative execution to steal application secrets pose a significant threat to modern computing systems. While program transformations can mitigate some Spectre attacks, more advanced attacks can divert control flow speculatively to bypass these protective instructions, rendering existing defenses useless. In this paper, we present Venkman: a system that employs program transformation to completely thwart Spectre attacks that poison entries in the Branch Target Buffer (BTB) and the Return Stack Buffer (RSB). Venkman transforms code so that all valid targets of a control-flow transfer have an identical alignment in the virtual address space; it further transforms all branches to ensure that all entries added to the BTB and RSB are properly aligned. By transforming all code this way, Venkman ensures that, in any program wanting Spectre defenses, all control-flow transfers, including speculative ones, do not skip over protective instructions Venkman adds to the code segment to mitigate Spectre attacks. Unlike existing defenses, Venkman does not reduce sharing of the BTB and RSB and does not flush these structures, allowing safe sharing and reuse among programs while maintaining strong protection against Spectre attacks. We built a prototype of Venkman on an IBM POWER8 machine. Our evaluation on the SPEC benchmarks and selected applications shows that Venkman increases execution time to 3.47$\times$ on average and increases code size to 1.94$\times$ on average when it is used to ensure that fences are executed to mitigate Spectre attacks. Our evaluation also shows that Spectre-resistant Software Fault Isolation (SFI) built using Venkman incurs a geometric mean of 2.42$\times$ space overhead and 1.68$\times$ performance overhead.

cs.CR↗
arXiv2019

An end-to-end Neural Network Framework for Text Clustering

The unsupervised text clustering is one of the major tasks in natural language processing (NLP) and remains a difficult and complex problem. Conventional \mbox{methods} generally treat this task using separated steps, including text representation learning and clustering the representations. As an improvement, neural methods have also been introduced for continuous representation learning to address the sparsity problem. However, the multi-step process still deviates from the unified optimization target. Especially the second step of cluster is generally performed with conventional methods such as k-Means. We propose a pure neural framework for text clustering in an end-to-end manner. It jointly learns the text representation and the clustering model. Our model works well when the context can be obtained, which is nearly always the case in the field of NLP. We have our method \mbox{evaluated} on two widely used benchmarks: IMDB movie reviews for sentiment classification and $20$-Newsgroup for topic categorization. Despite its simplicity, experiments show the model outperforms previous clustering methods by a large margin. Furthermore, the model is also verified on English wiki dataset as a large corpus.

cs.CL↗
arXiv2019

Cellular reproduction number, generation time and growth rate differ between human- and avian-adapted influenza strains

When analysing in vitro data, growth kinetics of influenza strains are often compared by computing their growth rates, which are sometimes used as proxies for fitness. However, analogous to mechanistic epidemic models, the growth rate can be defined as a function of two parameters: the basic reproduction number (the average number of cells each infected cell infects) and the mean generation time (the average length of a replication cycle). Using a mechanistic model, previously published data from experiments in human lung cells, and newly generated data, we compared estimates of all three parameters for six influenza A strains. Using previously published data, we found that the two human-adapted strains (pre-2009 seasonal H1N1, and pandemic H1N1) had a lower basic reproduction number, shorter mean generation time and slower growth rate than the two avian-adapted strains (H5N1 and H7N9). These same differences were then observed in data from new experiments where two strains were engineered to have different internal proteins (pandemic H1N1 and H5N1), but the same surface proteins (PR8), confirming our initial findings and implying that differences between strains were driven by internal genes. Also, the model predicted that the human-adapted strains underwent more replication cycles than the avian-adapted strains by the time of peak viral load, potentially accumulating mutations more quickly. These results suggest that the in vitro reproduction number, generation time and growth rate differ between human-adapted and avian-adapted influenza strains, and thus could be used to assess host adaptation of internal proteins to inform pandemic risk assessment.

q-bio.PE↗
arXiv2019

Element Replacement Approach by Reaction with Lewis Acidic Molten Salts to Synthesize Nanolaminated MAX Phases and MXenes

Nanolaminated materials are important because of their exceptional properties and wide range of applications. Here, we demonstrate a general approach to synthesize a series of Zn-based MAX phases and Cl-terminated MXenes originating from the replacement reaction between the MAX phase and the late transition metal halides. The approach is a top-down route that enables the late transitional element atom (Zn in the present case) to occupy the A site in the pre-existing MAX phase structure. Using this replacement reaction between Zn element from molten ZnCl2 and Al element in MAX phase precursors (Ti3AlC2, Ti2AlC, Ti2AlN, and V2AlC), novel MAX phases Ti3ZnC2, Ti2ZnC, Ti2ZnN, and V2ZnC were synthesized. When employing excess ZnCl2, Cl terminated MXenes (such as Ti3C2Cl2 and Ti2CCl2) were derived by a subsequent exfoliation of Ti3ZnC2 and Ti2ZnC due to the strong Lewis acidity of molten ZnCl2. These results indicate that A-site element replacement in traditional MAX phases by late transition metal halides opens the door to explore MAX phases that are not thermodynamically stable at high temperature and would be difficult to synthesize through the commonly employed powder metallurgy approach. In addition, this is the first time that exclusively Cl-terminated MXenes were obtained, and the etching effect of Lewis acid in molten salts provides a green and viable route to prepare MXenes through an HF-free chemical approach.

cond-mat.mtrl-sci↗
arXiv2019

COIN: A Large-scale Dataset for Comprehensive Instructional Video Analysis

There are substantial instructional videos on the Internet, which enables us to acquire knowledge for completing various tasks. However, most existing datasets for instructional video analysis have the limitations in diversity and scale,which makes them far from many real-world applications where more diverse activities occur. Moreover, it still remains a great challenge to organize and harness such data. To address these problems, we introduce a large-scale dataset called "COIN" for COmprehensive INstructional video analysis. Organized with a hierarchical structure, the COIN dataset contains 11,827 videos of 180 tasks in 12 domains (e.g., vehicles, gadgets, etc.) related to our daily life. With a new developed toolbox, all the videos are annotated effectively with a series of step descriptions and the corresponding temporal boundaries. Furthermore, we propose a simple yet effective method to capture the dependencies among different steps, which can be easily plugged into conventional proposal-based action detection methods for localizing important steps in instructional videos. In order to provide a benchmark for instructional video analysis, we evaluate plenty of approaches on the COIN dataset under different evaluation criteria. We expect the introduction of the COIN dataset will promote the future in-depth research on instructional video analysis for the community.

cs.CV↗
arXiv2019

Option Comparison Network for Multiple-choice Reading Comprehension

Multiple-choice reading comprehension (MCRC) is the task of selecting the correct answer from multiple options given a question and an article. Existing MCRC models typically either read each option independently or compute a fixed-length representation for each option before comparing them. However, humans typically compare the options at multiple-granularity level before reading the article in detail to make reasoning more efficient. Mimicking humans, we propose an option comparison network (OCN) for MCRC which compares options at word-level to better identify their correlations to help reasoning. Specially, each option is encoded into a vector sequence using a skimmer to retain fine-grained information as much as possible. An attention mechanism is leveraged to compare these sequences vector-by-vector to identify more subtle correlations between options, which is potentially valuable for reasoning. Experimental results on the human English exam MCRC dataset RACE show that our model outperforms existing methods significantly. Moreover, it is also the first model that surpasses Amazon Mechanical Turker performance on the whole dataset.

cs.CL↗
arXiv2018

Local density matrices of many-body states in the constant weight subspaces

Let $V=\bigotimes_{k=1}^{N} V_{k}$ be the $N$ spin-$j$ Hilbert space with $d=2j+1$-dimensional single particle space. We fix an orthonormal basis $\{|m_i\rangle\}$ for each $V_{k}$, with weight $m_i\in \{-j,\ldots j\}$. Let $V_{(w)}$ be the subspace of $V$ with a constant weight $w$, with an orthonormal basis $\{|m_1,\ldots,m_N\rangle\}$ subject to $\sum_k m_k=w$. We show that the combinatorial properties of the constant weight condition imposes strong constraints on the reduced density matrices for any vector $|ψ\rangle$ in the constant weight subspace, which limits the possible entanglement structures of $|ψ\rangle$. Our results find applications in the overlapping quantum marginal problems, quantum error-correcting codes, and the spin-network structures in quantum gravity.

quant-ph↗
arXiv2018

General Hardy-Type Paradox Based on Bell inequality and its Experimental Test

Local realistic models cannot completely describe all predictions of quantum mechanics. This is known as Bell's theorem that can be revealed either by violations of Bell inequality, or all-versus-nothing proof of nonlocality. Hardy's paradox is an important all-versus-nothing proof and is considered as "the simplest form of Bell's theorem". In this work, we theoretically build the general framework of Hardy-type paradox based on Bell inequality. Previous Hardy's paradoxes have been found to be special cases within the framework. Stronger Hardy-type paradox has been found even for the two-qubit two-setting case, and the corresponding successful probability is about four times larger than the original one, thus providing a more friendly test for experiment. We also find that GHZ paradox can be viewed as a perfect Hardy-type paradox. Meanwhile, we experimentally test the stronger Hardy-type paradoxes in a two-qubit system. Within the experimental errors, the experimental results coincide with the theoretical predictions.

quant-ph↗
arXiv2018

Deriving Einstein-Podolsky-Rosen steering inequalities from the few-body Abner Shimony inequalities

For the Abner Shimony (AS) inequalities, the simplest unified forms of directions attaining the maximum quantum violation are investigated. Based on these directions, a family of Einstein-Podolsky-Rosen (EPR) steering inequalities is derived from the AS inequalities in a systematic manner. For these inequalities, the local hidden state (LHS) bounds are strictly less than the local hidden variable (LHV) bounds. This means that the EPR steering is a form of quantum nonlocality strictly weaker than Bell-nonlocality.

quant-ph↗
arXiv2018

An Improved Evaluation Framework for Generative Adversarial Networks

In this paper, we propose an improved quantitative evaluation framework for Generative Adversarial Networks (GANs) on generating domain-specific images, where we improve conventional evaluation methods on two levels: the feature representation and the evaluation metric. Unlike most existing evaluation frameworks which transfer the representation of ImageNet inception model to map images onto the feature space, our framework uses a specialized encoder to acquire fine-grained domain-specific representation. Moreover, for datasets with multiple classes, we propose Class-Aware Frechet Distance (CAFD), which employs a Gaussian mixture model on the feature space to better fit the multi-manifold feature distribution. Experiments and analysis on both the feature level and the image level were conducted to demonstrate improvements of our proposed framework over the recently proposed state-of-the-art FID method. To our best knowledge, we are the first to provide counter examples where FID gives inconsistent results with human judgments. It is shown in the experiments that our framework is able to overcome the shortness of FID and improves robustness. Code will be made available.

cs.CV↗
arXiv2018

Chained Einstein-Podolsky-Rosen steering inequalities with improved visibility

It is known that the linear n-setting steering inequalities introduced in Ref. [Nature Phys. 6, 845 (2010)] are very efficient inequalities in detecting steerability of the Werner states by using optimal measurement axes. Here, we construct chained steering inequalities that have improved visibility for the Werner states under a finite number of settings. Specifically, the threshold values of quantum violation of our inequalities for the n=4,6,10 settings are lower than those of the linear steering inequalities. Furthermore, for almost all generalized Werner states, the chained steering inequalities always have improved visibility in comparison with the linear steering inequalities.

quant-ph↗
arXiv2018

LG/CY Correspondence for Elliptic Orbifold Curves via Modularity

We prove the Landau-Ginzburg/Calabi-Yau correspondence between the Gromov-Witten theory of each elliptic orbifold curve and its Fan-Jarvis-Ruan-Witten theory counterpart via modularity. We show that the correlation functions in these two enumerative theories are different representations of the same set of quasi-modular forms, expanded around different points on the upper-half plane. We relate these two representations by the Cayley transform.

math.AG↗
arXiv2017

Quantum Spacetime on a Quantum Simulator

We experimentally simulate the spin networks -- a fundamental description of quantum spacetime at the Planck level. We achieve this by simulating quantum tetrahedra and their interactions. The tensor product of these quantum tetrahedra comprises spin networks. In this initial attempt to study quantum spacetime by quantum information processing, on a four-qubit nuclear magnetic resonance quantum simulator, we simulate the basic module -- comprising five quantum tetrahedra -- of the interactions of quantum spacetime. By measuring the geometric properties on the corresponding quantum tetrahedra and simulate their interactions, our experiment serves as the basic module that represents the Feynman diagram vertex in the spin-network formulation of quantum spacetime.

quant-ph↗
arXiv2017

Genus one GW invariants of quintic threefolds via MSP localization

The moduli stack of Mixed Spin P-fields (MSP) provides an effective algorithm to evaluate all genus Gromov-Witten invariants of quintic Calabi-Yau threefolds. This paper is to apply the algorithm in genus one case. We use the localization formula, the proposed algorithm in [CLLL1, CLLL2], and Zinger's packaging technique to compute the genus one Gromov-Witten invariants of quintic Calabi-Yau threefolds. New hypergeometric series identities are also discovered in the process.

math.AG↗
arXiv2017

Revisiting the Effectiveness of Off-the-shelf Temporal Modeling Approaches for Large-scale Video Classification

This paper describes our solution for the video recognition task of ActivityNet Kinetics challenge that ranked the 1st place. Most of existing state-of-the-art video recognition approaches are in favor of an end-to-end pipeline. One exception is the framework of DevNet. The merit of DevNet is that they first use the video data to learn a network (i.e. fine-tuning or training from scratch). Instead of directly using the end-to-end classification scores (e.g. softmax scores), they extract the features from the learned network and then fed them into the off-the-shelf machine learning models to conduct video classification. However, the effectiveness of this line work has long-term been ignored and underestimated. In this submission, we extensively use this strategy. Particularly, we investigate four temporal modeling approaches using the learned features: Multi-group Shifting Attention Network, Temporal Xception Network, Multi-stream sequence Model and Fast-Forward Sequence Model. Experiment results on the challenging Kinetics dataset demonstrate that our proposed temporal modeling approaches can significantly improve existing approaches in the large-scale video recognition tasks. Most remarkably, our best single Multi-group Shifting Attention Network can achieve 77.7% in term of top-1 accuracy and 93.2% in term of top-5 accuracy on the validation set.

cs.CV↗
arXiv2017

Temporal Modeling Approaches for Large-scale Youtube-8M Video Understanding

This paper describes our solution for the video recognition task of the Google Cloud and YouTube-8M Video Understanding Challenge that ranked the 3rd place. Because the challenge provides pre-extracted visual and audio features instead of the raw videos, we mainly investigate various temporal modeling approaches to aggregate the frame-level features for multi-label video recognition. Our system contains three major components: two-stream sequence model, fast-forward sequence model and temporal residual neural networks. Experiment results on the challenging Youtube-8M dataset demonstrate that our proposed temporal modeling approaches can significantly improve existing temporal modeling approaches in the large-scale video recognition tasks. To be noted, our fast-forward LSTM with a depth of 7 layers achieves 82.75% in term of GAP@20 on the Kaggle Public test set.

cs.CV↗