arXiv subjects

Hong Liu

Publications and source records attributed to Hong Liu.

At least 379 records · Page 21Linked to original sources

arXiv2020

A dynamical mechanism for the Page curve from quantum chaos

If the evaporation of a black hole formed from a pure state is unitary, the entanglement entropy of the Hawking radiation should follow the Page curve, increasing from zero until near the halfway point of the evaporation, and then decreasing back to zero. The general argument for the Page curve is based on the assumption that the quantum state of the black hole plus radiation during the evaporation process is typical. In this paper, we show that the Page curve can result from a simple dynamical input in the evolution of the black hole, based on a recently proposed signature of quantum chaos, without resorting to typicality. Our argument is based on what we refer to as the "operator gas" approach, which allows one to understand the evolution of the microstate of the black hole from generic features of the Heisenberg evolution of operators. One key feature which leads to the Page curve is the possibility of dynamical processes where operators in the "gas" can "jump" outside the black hole, which we refer to as void formation processes. Such processes are initially exponentially suppressed, but dominate after a certain time scale, which can be used as a dynamical definition of the Page time. In the Hayden-Preskill protocol for young and old black holes, we show that void formation is also responsible for the transfer of information from the black hole to the radiation. We conjecture that void formation may provide a microscopic explanation for the recent semi-classical prescription of including islands in the calculation of the entanglement entropy of the radiation.

hep-th↗

arXiv2020

Guided learning for weakly-labeled semi-supervised sound event detection

We propose a simple but efficient method termed Guided Learning for weakly-labeled semi-supervised sound event detection (SED). There are two sub-targets implied in weakly-labeled SED: audio tagging and boundary detection. Instead of designing a single model by considering a trade-off between the two sub-targets, we design a teacher model aiming at audio tagging to guide a student model aiming at boundary detection to learn using the unlabeled data. The guidance is guaranteed by the audio tagging performance gap of the two models. In the meantime, the student model liberated from the trade-off is able to provide more excellent boundary detection results. We propose a principle to design such two models based on the relation between the temporal compression scale and the two sub-targets. We also propose an end-to-end semi-supervised learning process for these two models to enable their abilities to rise alternately. Experiments on the DCASE2018 Task4 dataset show that our approach achieves competitive performance.

cs.LG↗

arXiv2020

Hadamard Matrix Guided Online Hashing

Online image hashing has attracted increasing research attention recently, which receives large-scale data in a streaming manner to update the hash functions on-the-fly. Its key challenge lies in the difficulty of balancing the learning timeliness and model accuracy. To this end, most works follow a supervised setting, i.e., using class labels to boost the hashing performance, which defects in two aspects: First, strong constraints, e.g., orthogonal or similarity preserving, are used, which however are typically relaxed and lead to large accuracy drop. Second, large amounts of training batches are required to learn the up-to-date hash functions, which largely increase the learning complexity. To handle the above challenges, a novel supervised online hashing scheme termed Hadamard Matrix Guided Online Hashing (HMOH) is proposed in this paper. Our key innovation lies in introducing Hadamard matrix, which is an orthogonal binary matrix built via Sylvester method. In particular, to release the need of strong constraints, we regard each column of Hadamard matrix as the target code for each class label, which by nature satisfies several desired properties of hashing codes. To accelerate the online training, LSH is first adopted to align the lengths of target code and to-be-learned binary code. We then treat the learning of hash functions as a set of binary classification problems to fit the assigned target code. Finally, extensive experiments demonstrate the superior accuracy and efficiency of the proposed method over various state-of-the-art methods. Codes are available at https://github.com/lmbxmu/mycode.

cs.IR↗

arXiv2020

Dynamical Phase Transition from Nonequilibrium Dynamics of Dark Solitons

By holographic duality, we identify a novel dynamical phase transition which results from the temperature dependence of non-equilibrium dynamics of dark solitons in a superfluid.For a non-equilibrium superfluid system with an initial density of dark solitons, there exists a critical temperature $T_d$,above which the system relaxes to equilibrium by producing sound waves, while below which it goes through an intermediate phase with a finite density of vortex-antivortex pairs. In particular, as $T_d$ is approached from below, the density of vortex pairs scales as $(T_d - T)^γ$ with the critical exponent $γ= 1/2$.

hep-th↗

arXiv2019

Void formation in operator growth, entanglement, and unitarity

The structure of the Heisenberg evolution of operators plays a key role in explaining diverse processes in quantum many-body systems. In this paper, we discuss a new universal feature of operator evolution: an operator can develop a void during its evolution, where its nontrivial parts become separated by a region of identity operators. Such processes are present in both integrable and chaotic systems, and are required by unitarity. We show that void formation has important implications for unitarity of entanglement growth and generation of mutual information and multipartite entanglement. We study explicitly the probability distributions of void formation in a number of unitary circuit models, and conjecture that in a quantum chaotic system the distribution is given by the one we find in random unitary circuits, which we refer to as the random void distribution. We also show that random unitary circuits lead to the same pattern of entanglement growth for multiple intervals as in (1+1)-dimensional holographic CFTs after a global quench, which can be used to argue that the random void distribution leads to maximal entanglement growth.

quant-ph↗

arXiv2019

An End-to-end Approach for Lexical Stress Detection based on Transformer

The dominant automatic lexical stress detection method is to split the utterance into syllable segments using phoneme sequence and their time-aligned boundaries. Then we extract features from syllable to use classification method to classify the lexical stress. However, we can't get very accurate time boundaries of each phoneme and we have to design some features in the syllable segments to classify the lexical stress. Therefore, we propose a end-to-end approach using sequence to sequence model of transformer to estimate lexical stress. For this, we train transformer model using feature sequence of audio and their phoneme sequence with lexical stress marks. During the recognition process, the recognized phoneme sequence is restricted according to the original standard phoneme sequence without lexical stress marks, but the lexical stress mark of each phoneme is not limited. We train the model in different subset of Librispeech and do lexical stress recognition in TIMIT and L2-ARCTIC dataset. For all subsets, the end-to-end model will perform better than the syllable segments classification method. Our method can achieve a 6.36% phoneme error rate on the TIMIT dataset, which exceeds the 7.2% error rate in other studies.

eess.AS↗

arXiv2019

Towards Understanding the Transferability of Deep Representations

Deep neural networks trained on a wide range of datasets demonstrate impressive transferability. Deep features appear general in that they are applicable to many datasets and tasks. Such property is in prevalent use in real-world applications. A neural network pretrained on large datasets, such as ImageNet, can significantly boost generalization and accelerate training if fine-tuned to a smaller target dataset. Despite its pervasiveness, few effort has been devoted to uncovering the reason of transferability in deep feature representations. This paper tries to understand transferability from the perspectives of improved generalization, optimization and the feasibility of transferability. We demonstrate that 1) Transferred models tend to find flatter minima, since their weight matrices stay close to the original flat region of pretrained parameters when transferred to a similar target dataset; 2) Transferred representations make the loss landscape more favorable with improved Lipschitzness, which accelerates and stabilizes training substantially. The improvement largely attributes to the fact that the principal component of gradient is suppressed in the pretrained parameters, thus stabilizing the magnitude of gradient in back-propagation. 3) The feasibility of transferability is related to the similarity of both input and label. And a surprising discovery is that the feasibility is also impacted by the training stages in that the transferability first increases during training, and then declines. We further provide a theoretical analysis to verify our observations.

cs.LG↗

arXiv2019

A degree sequence Komlós theorem

An important result of Komlós [Tiling Turán theorems, Combinatorica, 2000] yields the asymptotically exact minimum degree threshold that ensures a graph $G$ contains an $H$-tiling covering an $x$th proportion of the vertices of $G$ (for any fixed $x \in (0,1)$ and graph $H$). We give a degree sequence strengthening of this result which allows for a large proportion of the vertices in the host graph $G$ to have degree substantially smaller than that required by Komlós' theorem. We also demonstrate that for certain graphs $H$, the degree sequence condition is essentially best possible in more than one sense.

math.CO↗

arXiv2019

Universal Perturbation Attack Against Image Retrieval

Universal adversarial perturbations (UAPs), a.k.a. input-agnostic perturbations, has been proved to exist and be able to fool cutting-edge deep learning models on most of the data samples. Existing UAP methods mainly focus on attacking image classification models. Nevertheless, little attention has been paid to attacking image retrieval systems. In this paper, we make the first attempt in attacking image retrieval systems. Concretely, image retrieval attack is to make the retrieval system return irrelevant images to the query at the top ranking list. It plays an important role to corrupt the neighbourhood relationships among features in image retrieval attack. To this end, we propose a novel method to generate retrieval-against UAP to break the neighbourhood relationships of image features via degrading the corresponding ranking metric. To expand the attack method to scenarios with varying input sizes or untouchable network parameters, a multi-scale random resizing scheme and a ranking distillation strategy are proposed. We evaluate the proposed method on four widely-used image retrieval datasets, and report a significant performance drop in terms of different metrics, such as mAP and mP@10. Finally, we test our attack methods on the real-world visual search engine, i.e., Google Images, which demonstrates the practical potentials of our methods.

cs.CV↗

arXiv2019

Guided Learning Convolution System for DCASE 2019 Task 4

In this paper, we describe in detail the system we submitted to DCASE2019 task 4: sound event detection (SED) in domestic environments. We employ a convolutional neural network (CNN) with an embedding-level attention pooling module to solve it. By considering the interference caused by the co-occurrence of multiple events in the unbalanced dataset, we utilize the disentangled feature to raise the performance of the model. To take advantage of the unlabeled data, we adopt Guided Learning for semi-supervised learning. A group of median filters with adaptive window sizes is utilized in the post-processing of output probabilities of the model. We also analyze the effect of the synthetic data on the performance of the model and finally achieve an event-based F-measure of 45.43% on the validation set and an event-based F-measure of 42.7% on the test set. The system we submitted to the challenge achieves the best performance compared to those of other participates.

eess.AS↗

arXiv2019

Regenesis and quantum traversable wormholes

Recent gravity discussions of a traversable wormhole indicate that in holographic systems signals generated by a source could reappear long after they have dissipated, with the need of only performing some simple operations. In this paper we argue the phenomenon, to which we refer as "regenesis", is universal in general quantum chaotic many-body systems, and elucidate its underlying physics. The essential elements behind the phenomenon are: (i) scrambling which in a chaotic system makes out-of-time-ordered correlation functions (OTOCs) vanish at large times; (ii) the entanglement structure of the state of the system. The latter aspect also implies that the regenesis phenomenon requires fine tuning of the initial state. Compared to other manifestations of quantum chaos such as the initial growth of OTOCs which deals with early times, and a random matrix-type energy spectrum which reflects very large time behavior, regenesis concerns with intermediate times, of order the scrambling time of a system. We also study the phenomenon in detail in general two-dimensional conformal field theories in the large central charge limit, and highlight some interesting features including a resonant enhancement of regenesis signals near the scrambling time and their oscillations in coupling. Finally, we discuss gravity implications of the phenomenon for systems with a gravity dual, arguing that there exist regimes for which traversability of a wormhole is quantum in nature, i.e. cannot be associated with a semi-classical spacetime causal structure.

hep-th↗

arXiv2019

Expectation-Maximization Attention Networks for Semantic Segmentation

Self-attention mechanism has been widely used for various tasks. It is designed to compute the representation of each position by a weighted sum of the features at all positions. Thus, it can capture long-range relations for computer vision tasks. However, it is computationally consuming. Since the attention maps are computed w.r.t all other positions. In this paper, we formulate the attention mechanism into an expectation-maximization manner and iteratively estimate a much more compact set of bases upon which the attention maps are computed. By a weighted summation upon these bases, the resulting representation is low-rank and deprecates noisy information from the input. The proposed Expectation-Maximization Attention (EMA) module is robust to the variance of input and is also friendly in memory and computation. Moreover, we set up the bases maintenance and normalization methods to stabilize its training procedure. We conduct extensive experiments on popular semantic segmentation benchmarks including PASCAL VOC, PASCAL Context and COCO Stuff, on which we set new records.

cs.CV↗

arXiv2019

Sample phase gradient and fringe phase shift in dual phase grating X-ray interferometry

One of the key tasks in grating based x-ray phase contrast imaging is to accurately retrieve local phase gradients of a sample from measured intensity fringe shifts. To fulfill this task in dual phase grating interferometry, one needs to know the exact mathematical relationship between the two. In this work, using intuitive analysis of the sample-generated fringe shifts based on the beat pattern formation mechanism, the authors derived the formulas relating sample's phase gradients to fringe phase shifts. These formulas provide also a design optimization tool for dual phase grating interferometry.

physics.med-ph↗

arXiv2019

Smart wing rotation and ingenious leading edge vortex control modulate the unconventional forces during insects flights

Compared with fixed-wing flight, flapping flight can generate a higher lift and is also more maneuverable, largely resulting from the benefits of wing rotation. By analyzing the real wing kinematics of fruit flies, we found that the wing rotation is smart modulated and can be simplified into the advanced $Φ>0$, symmetrical $Φ=0$, and delayed $Φ<0$ modes, where $Φ$ is the phase between stroke reversal and wing rotation. Thus, an experiment of a Robotic fly at Reynolds number of 240 was designed to address how insects control the flow patterns to modulate the flight forces by the smart wing rotation. Experimental results demonstrated that the flow pattern during a half of stroke is featured by the formation of a spiral LEV, which has a convex shape and connects to the trailing vortex via a tip vortex and a root vortex to form a vortex loop. Through wing rotation, the formation of LEV is enhanced and the pinch-off of LEV can be significantly delayed, resulting in the stabilization of LEV and the generation of unconventional forces. In addition, by modulating the modes of wing rotation, insects obtain distinct wing-vortex patterns including positive, neutral, and negative camber patterns, by which insects control the magnitude and direction of unconventional forces owing to the wake capture mechanism. Therefore, the smart wing rotation not only contributes to the generation of unconventional forces, but also accounts for the control of flow pattern during insects steering maneuvers.

physics.flu-dyn↗

arXiv2019

Identifying Illicit Accounts in Large Scale E-payment Networks -- A Graph Representation Learning Approach

Rapid and massive adoption of mobile/ online payment services has brought new challenges to the service providers as well as regulators in safeguarding the proper uses such services/ systems. In this paper, we leverage recent advances in deep-neural-network-based graph representation learning to detect abnormal/ suspicious financial transactions in real-world e-payment networks. In particular, we propose an end-to-end Graph Convolution Network (GCN)-based algorithm to learn the embeddings of the nodes and edges of a large-scale time-evolving graph. In the context of e-payment transaction graphs, the resultant node and edge embeddings can effectively characterize the user-background as well as the financial transaction patterns of individual account holders. As such, we can use the graph embedding results to drive downstream graph mining tasks such as node-classification to identify illicit accounts within the payment networks. Our algorithm outperforms state-of-the-art schemes including GraphSAGE, Gradient Boosting Decision Tree and Random Forest to deliver considerably higher accuracy (94.62% and 86.98% respectively) in classifying user accounts within 2 practical e-payment transaction datasets. It also achieves outstanding accuracy (97.43%) for another biomedical entity identification task while using only edge-related information.

cs.SI↗

arXiv2019

Structure and Supersaturation for Intersecting Families

The extremal problems regarding the maximum possible size of intersecting families of various combinatorial objects have been extensively studied. In this paper, we investigate supersaturation extensions, which in this context ask for the minimum number of disjoint pairs that must appear in families larger than the extremal threshold. We study the minimum number of disjoint pairs in families of permutations and in $k$-uniform set families, and determine the structure of the optimal families. Our main tool is a removal lemma for disjoint pairs. We also determine the typical structure of $k$-uniform set families without matchings of size $s$ when $n \ge 2sk + 38s^4$, and show that almost all $k$-uniform intersecting families on vertex set $[n]$ are trivial when $n\ge (2+o(1))k$.

math.CO↗

arXiv2019

Clarification on Generalized Lau condition for X-ray interferometers based on dual phase gratings

To implement dual phase grating x-ray interferometry with x-ray tubes, one needs to incorporate an absorbing source grating. In order to attain good fringe visibility, the period of a source grating should be subject to a stringent condition. In literature some authors claim that the Lau-condition in Talbot-Lau interferometry can be literally transferred to dual phase grating interferometry. In this work we show that this statement in literature is incorrect. Instead, through an intuitive geometrical analysis of fringe formation, we derived a new generalized Lau-condition that provides a useful design tool for implementation of dual phase grating interferometry.

physics.med-ph↗

arXiv2019

Modular Flow of Excited States

We develop new techniques for studying the modular and the relative modular flows of general excited states. We show that the class of states obtained by acting on the vacuum (or any cyclic and separating state) with invertible operators from the algebra of a region is dense in the Hilbert space. This enables us to express the modular and the relative modular operators, as well as the relative entropies of generic excited states in terms of the vacuum modular operator and the operator that creates the state. In particular, the modular and the relative modular flows of any state can be expanded in terms of the modular flow of operators in vacuum. We illustrate the formalism with simple examples including states close to the vacuum, and coherent and squeezed states in generalized free field theory.

hep-th↗