arXiv subjects
Publications and source records attributed to Liang Zhang.
We would like to withdraw this article for the following reasons: 1 this article is not satisfactory for limited language and theoretical description; 2 we have enriched and revised this article with the help of other authors; 3 we must update the author contribution information.
Multimodal processing has attracted much attention lately especially with the success of pre-training. However, the exploration has mainly focused on vision-language pre-training, as introducing more modalities can greatly complicate model design and optimization. In this paper, we extend the stateof-the-art Vision-Language model CLIP to accommodate the audio modality for Vision-Language-Audio multimodal processing. Specifically, we apply inter-modal and intra-modal contrastive learning to explore the correlation between audio and other modalities in addition to the inner characteristics of the audio modality. Moreover, we further design an audio type token to dynamically learn different audio information type for different scenarios, as both verbal and nonverbal heterogeneous information is conveyed in general audios. Our proposed CLIP4VLA model is validated in different downstream tasks including video retrieval and video captioning, and achieves the state-of-the-art performance on the benchmark datasets of MSR-VTT, VATEX, and Audiocaps.
MAXI J1535-571 is a black-hole X-ray binary that in 2017 exhibited a very bright outburst which reached a peak flux of up to 5 Crab in the 2-20 keV band. Given the high flux, several X-ray space observatories obtained unprecedented high signal-to-noise data of key parts of the outburst. In our previous paper we studied the corona of MAXI J1535-571 in the hard-intermediate state (HIMS) with Insight-HXMT. In this paper we focus on the study of the corona in the soft-intermediate state (SIMS) through the spectral-timing analysis of 26 NICER detections of the type-B quasi-periodic oscillations (QPOs). From simultaneous fits of the energy, rms and lag spectra of these QPOs with our time-dependent Comptonization model, we find that in the SIMS the corona size is ~ 6500 km and vertically extended. We detect a narrow iron line in the energy spectra, which we interpret to be due to the illumination of the outer part of the accretion disk by this large corona. We follow the evolution of the corona and the radio jet during the HIMS-SIMS transition, and find that the jet flux peaks after the time when the corona extends to its maximum vertical size. The jet flux starts to decay after the corona contracts vertically towards the black hole. This behavior points to a connection between the X-ray corona and the radio jet similar to that seen in other sources.
The artificial intelligence (AI) system has achieved expert-level performance in electrocardiogram (ECG) signal analysis. However, in underdeveloped countries or regions where the healthcare information system is imperfect, only paper ECGs can be provided. Analysis of real-world ECG images (photos or scans of paper ECGs) remains challenging due to complex environments or interference. In this study, we present an AI system developed to detect and screen cardiac abnormalities (CAs) from real-world ECG images. The system was evaluated on a large dataset of 52,357 patients from multiple regions and populations across the world. On the detection task, the AI system obtained area under the receiver operating curve (AUC) of 0.996 (hold-out test), 0.994 (external test 1), 0.984 (external test 2), and 0.979 (external test 3), respectively. Meanwhile, the detection results of AI system showed a strong correlation with the diagnosis of cardiologists (cardiologist 1 (R=0.794, p<1e-3), cardiologist 2 (R=0.812, p<1e-3)). On the screening task, the AI system achieved AUCs of 0.894 (hold-out test) and 0.850 (external test). The screening performance of the AI system was better than that of the cardiologists (AI system (0.846) vs. cardiologist 1 (0.520) vs. cardiologist 2 (0.480)). Our study demonstrates the feasibility of an accurate, objective, easy-to-use, fast, and low-cost AI system for CA detection and screening. The system has the potential to be used by healthcare professionals, caregivers, and general users to assess CAs based on real-world ECG images.
This paper takes an initial step to systematically investigate the generalization bounds of algorithms for solving nonconvex-(strongly)-concave (NC-SC/NC-C) stochastic minimax optimization measured by the stationarity of primal functions. We first establish algorithm-agnostic generalization bounds via uniform convergence between the empirical minimax problem and the population minimax problem. The sample complexities for achieving $ε$-generalization are $\tilde{\mathcal{O}}(dκ^2ε^{-2})$ and $\tilde{\mathcal{O}}(dε^{-4})$ for NC-SC and NC-C settings, respectively, where $d$ is the dimension and $κ$ is the condition number. We further study the algorithm-dependent generalization bounds via stability arguments of algorithms. In particular, we introduce a novel stability notion for minimax problems and build a connection between generalization bounds and the stability notion. As a result, we establish algorithm-dependent generalization bounds for stochastic gradient descent ascent (SGDA) algorithm and the more general sampling-determined algorithms.
We present a detailed spectral and temporal analysis of the black-hole candidate MAXI~J1535$-$571 using NICER observations in September and October 2017. We focus specifically on observations in the hard-intermediate state when the source shows type-C quasi-periodic oscillations (QPOs). We fitted the time-averaged spectrum of the source and the rms and phase-lag spectra of the QPO with a one-component time-dependent Comptonization model. We found that the corona contracts from $\sim 10^4$ to $\sim 3 \times 10^3$ km as the QPO frequency increases from $\sim 1.8$ Hz to $\sim 9.0$ Hz. The fits suggest that the system would consists of two coronas, a small one that dominates the time-averaged spectrum and a larger one, possibly the jet, that dominates the rms and lag spectra of the QPO. We found a significant break in the relation of the spectral parameters of the source and the properties of the QPO, including its lag spectra, with QPO frequency. The change in the relations happens when the QPO frequency crosses a critical frequency $ν_c \approx 3.0$ Hz. Interestingly, the QPO reaches this critical frequency simultaneously as the radio emission from the jet in this source is quenched.
Magnetic flux ropes (MFRs), sets of coherently twisted magnetic field lines, are believed as core structures of various solar eruptions. Their evolution plays an important role to understand the physical mechanisms of solar eruptions, and can shed light on adverse space weather near the Earth. However, the erupting MFRs are occasionally prevented by strong overlying magnetic fields, and the MFR evolution during the descending phase in the confined cases is lack of attention. Here, we present the deformation of an erupting MFR accompanied by a confined double-peaked solar flare. The first peak corresponded to the MFR eruption in a standard flare model, and the second peak was closely associated with the flashings of an underlying sheared arcade (SA), the reversal slipping motion of the L-shaped flare ribbon, the falling of the MFR, and the shifting of top of filament threads. All results suggest that the confined MFR eruption involved in two-step magnetic reconnection presenting two distinct episodes of energy release in the flare impulsive phase, and the latter magnetic reconnection between the confined MFR and the underlying SA caused the deformation of MFR.
Document-level relation extraction (RE) aims to extract the relations between entities from the input document that usually containing many difficultly-predicted entity pairs whose relations can only be predicted through relational inference. Existing methods usually directly predict the relations of all entity pairs of input document in a one-pass manner, ignoring the fact that predictions of some entity pairs heavily depend on the predicted results of other pairs. To deal with this issue, in this paper, we propose a novel document-level RE model with iterative inference. Our model is mainly composed of two modules: 1) a base module expected to provide preliminary relation predictions on entity pairs; 2) an inference module introduced to refine these preliminary predictions by iteratively dealing with difficultly-predicted entity pairs depending on other pairs in an easy-to-hard manner. Unlike previous methods which only consider feature information of entity pairs, our inference module is equipped with two Extended Cross Attention units, allowing it to exploit both feature information and previous predictions of entity pairs during relational inference. Furthermore, we adopt a two-stage strategy to train our model. At the first stage, we only train our base module. During the second stage, we train the whole model, where contrastive learning is introduced to enhance the training of inference module. Experimental results on three commonly-used datasets show that our model consistently outperforms other competitive baselines.
In time-domain astronomy, a substantial number of transients will be discovered by multi-wavelength and multi-messenger observatories, posing a great challenge for follow-up capabilities. We have thus proposed an intelligent X-ray constellation, the Chasing All Transients Constellation Hunters (CATCH) space mission. Consisting of 126 micro-satellites in three types, CATCH will have the capability to perform follow-up observations for a large number of different types of transients simultaneously. Each satellite in the constellation will carry lightweight X-ray optics and use a deployable mast to increase the focal length. The combination of different optics and detector systems enables different types of satellites to have multiform observation capabilities, including timing, spectroscopy, imaging, and polarization. Controlled by the intelligent system, different satellites can cooperate to perform uninterrupted monitoring, all-sky follow-up observations, and scanning observations with a flexible field of view (FOV) and multi-dimensional observations. Therefore, CATCH will be a powerful mission to study the dynamic universe. Here, we present the current design of the spacecraft, optics, detector system, constellation configuration and observing modes, as well as the development plan.
Unsupervised region representation learning aims to extract dense and effective features from unlabeled urban data. While some efforts have been made for solving this problem based on multiple views, existing methods are still insufficient in extracting representations in a view and/or incorporating representations from different views. Motivated by the success of contrastive learning for representation learning, we propose to leverage it for multi-view region representation learning and design a model called ReMVC (Region Embedding with Multi-View Contrastive Learning) by following two guidelines: i) comparing a region with others within each view for effective representation extraction and ii) comparing a region with itself across different views for cross-view information sharing. We design the intra-view contrastive learning module which helps to learn distinguished region embeddings and the inter-view contrastive learning module which serves as a soft co-regularizer to constrain the embedding parameters and transfer knowledge across multi-views. We exploit the learned region embeddings in two downstream tasks named land usage clustering and region popularity prediction. Extensive experiments demonstrate that our model achieves impressive improvements compared with seven state-of-the-art baseline methods, and the margins are over 30% in the land usage clustering task.
When performing complex tasks, humans naturally reason at multiple temporal and spatial resolutions simultaneously. We contend that for an artificially intelligent agent to effectively model human teammates, i.e., demonstrate computational theory of mind (ToM), it should do the same. In this paper, we present an approach for integrating high and low-resolution spatial and temporal information to predict human behavior in real time and evaluate it on data collected from human subjects performing simulated urban search and rescue (USAR) missions in a Minecraft-based environment. Our model composes neural networks for high and low-resolution feature extraction with a neural network for behavior prediction, with all three networks trained simultaneously. The high-resolution extractor encodes dynamically changing goals robustly by taking as input the Manhattan distance difference between the humans' Minecraft avatars and candidate goals in the environment for the latest few actions, computed from a high-resolution gridworld representation. In contrast, the low-resolution extractor encodes participants' historical behavior using a historical state matrix computed from a low-resolution graph representation. Through supervised learning, our model acquires a robust prior for human behavior prediction, and can effectively deal with long-term observations. Our experimental results demonstrate that our method significantly improves prediction accuracy compared to approaches that only use high-resolution information.
Human decision-making often involves combining similar states into categories and reasoning at the level of the categories rather than the actual states. Guided by this intuition, we propose a novel method for clustering state features in deep reinforcement learning (RL) methods to improve their interpretability. Specifically, we propose a plug-and-play framework termed \emph{vector quantized reinforcement learning} (VQ-RL) that extends classic RL pipelines with an auxiliary classification task based on vector quantized (VQ) encoding and aligns with policy training. The VQ encoding method categorizes features with similar semantics into clusters and results in tighter clusters with better separation compared to classic deep RL methods, thus enabling neural models to learn similarities and differences between states better. Furthermore, we introduce two regularization methods to help increase the separation between clusters and avoid the risks associated with VQ training. In simulations, we demonstrate that VQ-RL improves interpretability and investigate its impact on robustness and generalization of deep RL.
Dynamic statistical process monitoring methods have been widely studied and applied in modern industrial processes. These methods aim to extract the most predictable temporal information and develop the corresponding dynamic monitoring schemes. However, measurement noise is widespread in real-world industrial processes, and ignoring its effect will lead to sub-optimal modeling and monitoring performance. In this article, a probabilistic predictable feature analysis (PPFA) is proposed for multivariate time series modeling, and a multi-step dynamic predictive monitoring scheme is developed. The model parameters are estimated with an efficient expectation-maximum algorithm, where the genetic algorithm and Kalman filter are designed and incorporated. Further, a novel dynamic statistical monitoring index, Dynamic Index, is proposed as an important supplement of $\text{T}^2$ and $\text{SPE}$ to detect dynamic anomalies. The effectiveness of the proposed algorithm is demonstrated via its application on the three-phase flow facility and a medium speed coal mill.
Fast radio bursts (FRBs) are short pulses observed in radio frequencies usually originating from cosmological distances. The discovery of FRB 200428 and its X-ray counterpart from the Galactic magnetar SGR J1935+2154 suggests that at least some FRBs can be generated by magnetars. However, the majority of X-ray bursts from magnetars are not associated with radio emission. The fact that only in rare cases can an FRB be generated raises the question regarding the special triggering mechanism of FRBs. Here we report a giant glitch from SGR J1935+2154, which occurred approximately $3.1\pm2.5$\,day before FRB 200428, with $Δν=19.8\pm1.4$ {\rm $μ$Hz} and $Δ\dotν=6.3\pm1.1$\,pHz s$^{-1}$. The corresponding spin-down power change rate $Δ\dotν/\dotν$ is among the largest in all the detected pulsar glitches. The glitch contains a delayed spin-up process that is only detected in the Crab pulsar and the magnetar 1E 2259+586, a large persistent offset of the spin-down rate, and a recovery component which is about one order of magnitude smaller than the persistent one. The temporal coincidence between the glitch and FRB 200428 suggests a physical connection between the two. The internally triggered giant glitch of the magnetar likely altered the magnetosphere structure dramatically in favour of FRB generation, which subsequently triggered many X-ray bursts and eventually FRB 200428 through additional crustal cracking and Alfvén wave excitation and propagation.
Using four mixed bivariate distributions (Normal distribution, Skew-Normal distribution, Student distribution, Skew-Student distribution) and bootstrap re-sampling analysis, we analyze the samples of CGRO/BATSE, Swift/BAT and Fermi/GBM gamma-ray bursts in detail on the t90-hardness ratio plane. The Bayesian information criterion is used to judge the goodness of fit for each sample, comprehensively. It is found that all the three samples show a symmetric (either normal or student) distribution. It is also found that the existence of three classes of gamma-ray bursts is preferred by the three samples, but the strength of this preference varies with the sample size: when the sample size of the data set is larger, the preference of three classes scheme becomes weaker. Therefore, the appearance of an intermediate class may be caused by a small sample size and the possibility that there are only two classes of gamma-ray bursts still cannot be expelled yet. A further bootstrap re-sampling analysis also confirms this result.
Many ncRNAs function through RNA-RNA interactions. Fast and reliable RNA structure prediction with consideration of RNA-RNA interaction is useful. Some existing tools are less accurate due to omitting the competing of intermolecular and intramolecular base pairs, or focus more on predicting the binding region rather than predicting the complete secondary structure of two interacting strands. Vienna RNAcofold, which reduces the problem into the classical single sequence folding by concatenating two strands, scales in cubic time against the combined sequence length, and is slow for long sequences. To address these issues, we present LinearCoFold, which predicts the complete minimum free energy structure of two strands in linear runtime, and LinearCoPartition, which calculates the cofolding partition function and base pairing probabilities in linear runtime. LinearCoFold and LinearCoPartition follows the concatenation strategy of RNAcofold, but are orders of magnitude faster than RNAcofold. For example, on a sequence pair with combined length of 26,190 nt, LinearCoFold is 86.8x faster than RNAcofold MFE mode (0.6 minutes vs. 52.1 minutes), and LinearCoPartition is 642.3x faster than RNAcofold partition function mode (1.8 minutes vs. 1156.2 minutes). Different from the local algorithms, LinearCoFold and LinearCoPartition are global cofolding algorithms without restriction on base pair length. Surprisingly, LinearCoFold and LinearCoPartition's predictions have higher PPV and sensitivity of intermolecular base pairs. Furthermore, we apply LinearCoFold to predict the RNA-RNA interaction between SARS-CoV-2 gRNA and human U4 snRNA, which has been experimentally studied, and observe that LinearCoFold's prediction correlates better to the wet lab results.
We study differentially private (DP) algorithms for smooth stochastic minimax optimization, with stochastic minimization as a byproduct. The holy grail of these settings is to guarantee the optimal trade-off between the privacy and the excess population loss, using an algorithm with a linear time-complexity in the number of training samples. We provide a general framework for solving differentially private stochastic minimax optimization (DP-SMO) problems, which enables the practitioners to bring their own base optimization algorithm and use it as a black-box to obtain the near-optimal privacy-loss trade-off. Our framework is inspired from the recently proposed Phased-ERM method [22] for nonsmooth differentially private stochastic convex optimization (DP-SCO), which exploits the stability of the empirical risk minimization (ERM) for the privacy guarantee. The flexibility of our approach enables us to sidestep the requirement that the base algorithm needs to have bounded sensitivity, and allows the use of sophisticated variance-reduced accelerated methods to achieve near-linear time-complexity. To the best of our knowledge, these are the first near-linear time algorithms with near-optimal guarantees on the population duality gap for smooth DP-SMO, when the objective is (strongly-)convex--(strongly-)concave. Additionally, based on our flexible framework, we enrich the family of near-linear time algorithms for smooth DP-SCO with the near-optimal privacy-loss trade-off.
Collaborative edge computing (CEC) is an emerging paradigm enabling sharing of the coupled data, computation, and networking resources among heterogeneous geo-distributed edge nodes. Recently, there has been a trend to orchestrate and schedule containerized application workloads in CEC, while Kubernetes has become the de-facto standard broadly adopted by the industry and academia. However, Kubernetes is not preferable for CEC because its design is not dedicated to edge computing and neglects the unique features of edge nativeness. More specifically, Kubernetes primarily ensures resource provision of workloads while neglecting the performance requirements of edge-native applications, such as throughput and latency. Furthermore, Kubernetes neglects the inner dependencies of edge-native applications and fails to consider data locality and networking resources, leading to inferior performance. In this work, we design and develop ENTS, the first edge-native task scheduling system, to manage the distributed edge resources and facilitate efficient task scheduling to optimize the performance of edge-native applications. ENTS extends Kubernetes with the unique ability to collaboratively schedule computation and networking resources by comprehensively considering job profile and resource status. We showcase the superior efficacy of ENTS with a case study on data streaming applications. We mathematically formulate a joint task allocation and flow scheduling problem that maximizes the job throughput. We design two novel online scheduling algorithms to optimally decide the task allocation, bandwidth allocation, and flow routing policies. The extensive experiments on a real-world edge video analytics application show that ENTS achieves 43\%-220\% higher average job throughput compared with the state-of-the-art.