arXiv subjects

Xinbing Wang

Publications and source records attributed to Xinbing Wang.

At least 109 records · Page 6Linked to original sources

arXiv2022

Syntax-guided Localized Self-attention by Constituency Syntactic Distance

Recent works have revealed that Transformers are implicitly learning the syntactic information in its lower layers from data, albeit is highly dependent on the quality and scale of the training data. However, learning syntactic information from data is not necessary if we can leverage an external syntactic parser, which provides better parsing quality with well-defined syntactic structures. This could potentially improve Transformer's performance and sample efficiency. In this work, we propose a syntax-guided localized self-attention for Transformer that allows directly incorporating grammar structures from an external constituency parser. It prohibits the attention mechanism to overweight the grammatically distant tokens over close ones. Experimental results show that our model could consistently improve translation performance on a variety of machine translation datasets, ranging from small to large dataset sizes, and with different source languages.

cs.CL↗

arXiv2022

RASAT: Integrating Relational Structures into Pretrained Seq2Seq Model for Text-to-SQL

Relational structures such as schema linking and schema encoding have been validated as a key component to qualitatively translating natural language into SQL queries. However, introducing these structural relations comes with prices: they often result in a specialized model structure, which largely prohibits using large pretrained models in text-to-SQL. To address this problem, we propose RASAT: a Transformer seq2seq architecture augmented with relation-aware self-attention that could leverage a variety of relational structures while inheriting the pretrained parameters from the T5 model effectively. Our model can incorporate almost all types of existing relations in the literature, and in addition, we propose introducing co-reference relations for the multi-turn scenario. Experimental results on three widely used text-to-SQL datasets, covering both single-turn and multi-turn scenarios, have shown that RASAT could achieve state-of-the-art results across all three benchmarks (75.5% EX on Spider, 52.6% IEX on SParC, and 37.4% IEX on CoSQL).

cs.CL↗

arXiv2022

KnowledgeShovel: An AI-in-the-Loop Document Annotation System for Scientific Knowledge Base Construction

Constructing a comprehensive, accurate, and useful scientific knowledge base is crucial for human researchers synthesizing scientific knowledge and for enabling Al-driven scientific discovery. However, the current process is difficult, error-prone, and laborious due to (1) the enormous amount of scientific literature available; (2) the highly-specialized scientific domains; (3) the diverse modalities of information (text, figure, table); and, (4) the silos of scientific knowledge in different publications with inconsistent formats and structures. Informed by a formative study and iterated with participatory design workshops, we designed and developed KnowledgeShovel, an Al-in-the-Loop document annotation system for researchers to construct scientific knowledge bases. The design of KnowledgeShovel introduces a multi-step multi-modal human-AI collaboration pipeline that aligns with users' existing workflows to improve data accuracy while reducing the human burden. A follow-up user evaluation with 7 geoscience researchers shows that KnowledgeShovel can enable efficient construction of scientific knowledge bases with satisfactory accuracy.

cs.DL↗

arXiv2022

IdeaReader: A Machine Reading System for Understanding the Idea Flow of Scientific Publications

Understanding the origin and influence of the publication's idea is critical to conducting scientific research. However, the proliferation of scientific publications makes it difficult for researchers to sort out the evolution of all relevant literature. To this end, we present IdeaReader, a machine reading system that finds out which papers are most likely to inspire or be influenced by the target publication and summarizes the ideas of these papers in natural language. Specifically, IdeaReader first clusters the references and citations (first-order or higher-order) of the target publication, and the obtained clusters are regarded as the topics that inspire or are influenced by the target publication. It then picks out the important papers from each cluster to extract the skeleton of the idea flow. Finally, IdeaReader automatically generates a literature review of the important papers in each topic. Our system can help researchers gain insight into how scientific ideas flow from the target publication's references to citations by the automatically generated survey and the visualization of idea flow.

cs.DL↗

arXiv2022

INFINITY: A Simple Yet Effective Unsupervised Framework for Graph-Text Mutual Conversion

Graph-to-text (G2T) generation and text-to-graph (T2G) triple extraction are two essential tasks for constructing and applying knowledge graphs. Existing unsupervised approaches turn out to be suitable candidates for jointly learning the two tasks due to their avoidance of using graph-text parallel data. However, they are composed of multiple modules and still require both entity information and relation type in the training process. To this end, we propose INFINITY, a simple yet effective unsupervised approach that does not require external annotation tools or additional parallel information. It achieves fully unsupervised graph-text mutual conversion for the first time. Specifically, INFINITY treats both G2T and T2G as a bidirectional sequence generation task by fine-tuning only one pretrained seq2seq model. A novel back-translation-based framework is then designed to automatically generate continuous synthetic parallel data. To obtain reasonable graph sequences with structural information from source texts, INFINITY employs reward-based training loss by leveraging the advantage of reward augmented maximum likelihood. As a fully unsupervised framework, INFINITY is empirically verified to outperform state-of-the-art baselines for G2T and T2G tasks.

cs.CL↗

arXiv2022

Disentangled Graph Contrastive Learning for Review-based Recommendation

User review data is helpful in alleviating the data sparsity problem in many recommender systems. In review-based recommendation methods, review data is considered as auxiliary information that can improve the quality of learned user/item or interaction representations for the user rating prediction task. However, these methods usually model user-item interactions in a holistic manner and neglect the entanglement of the latent factors behind them, e.g., price, quality, or appearance, resulting in suboptimal representations and reducing interpretability. In this paper, we propose a Disentangled Graph Contrastive Learning framework for Review-based recommendation (DGCLR), to separately model the user-item interactions based on different latent factors through the textual review data. To this end, we first model the distributions of interactions over latent factors from both semantic information in review data and structural information in user-item graph data, forming several factor graphs. Then a factorized message passing mechanism is designed to learn disentangled user/item representations on the factor graphs, which enable us to further characterize the interactions and adaptively combine the predicted ratings from multiple factors via a devised attention mechanism. Finally, we set two factor-wise contrastive learning objectives to alleviate the sparsity issue and model the user/item and interaction features pertinent to each factor more accurately. Empirical results over five benchmark datasets validate the superiority of DGCLR over the state-of-the-art methods. Further analysis is offered to interpret the learned intent factors and rating prediction in DGCLR.

cs.IR↗

arXiv2022

Geometer: Graph Few-Shot Class-Incremental Learning via Prototype Representation

With the tremendous expansion of graphs data, node classification shows its great importance in many real-world applications. Existing graph neural network based methods mainly focus on classifying unlabeled nodes within fixed classes with abundant labeling. However, in many practical scenarios, graph evolves with emergence of new nodes and edges. Novel classes appear incrementally along with few labeling due to its newly emergence or lack of exploration. In this paper, we focus on this challenging but practical graph few-shot class-incremental learning (GFSCIL) problem and propose a novel method called Geometer. Instead of replacing and retraining the fully connected neural network classifer, Geometer predicts the label of a node by finding the nearest class prototype. Prototype is a vector representing a class in the metric space. With the pop-up of novel classes, Geometer learns and adjusts the attention-based prototypes by observing the geometric proximity, uniformity and separability. Teacher-student knowledge distillation and biased sampling are further introduced to mitigate catastrophic forgetting and unbalanced labeling problem respectively. Experimental results on four public datasets demonstrate that Geometer achieves a substantial improvement of 9.46% to 27.60% over state-of-the-art methods.

cs.AI↗

arXiv2022

Spatio-Temporal Graph Few-Shot Learning with Cross-City Knowledge Transfer

Spatio-temporal graph learning is a key method for urban computing tasks, such as traffic flow, taxi demand and air quality forecasting. Due to the high cost of data collection, some developing cities have few available data, which makes it infeasible to train a well-performed model. To address this challenge, cross-city knowledge transfer has shown its promise, where the model learned from data-sufficient cities is leveraged to benefit the learning process of data-scarce cities. However, the spatio-temporal graphs among different cities show irregular structures and varied features, which limits the feasibility of existing Few-Shot Learning (\emph{FSL}) methods. Therefore, we propose a model-agnostic few-shot learning framework for spatio-temporal graph called ST-GFSL. Specifically, to enhance feature extraction by transfering cross-city knowledge, ST-GFSL proposes to generate non-shared parameters based on node-level meta knowledge. The nodes in target city transfer the knowledge via parameter matching, retrieving from similar spatio-temporal characteristics. Furthermore, we propose to reconstruct the graph structure during meta-learning. The graph reconstruction loss is defined to guide structure-aware learning, avoiding structure deviation among different datasets. We conduct comprehensive experiments on four traffic speed prediction benchmarks and the results demonstrate the effectiveness of ST-GFSL compared with state-of-the-art methods.

cs.LG↗

arXiv2022

Leveraging Unimodal Self-Supervised Learning for Multimodal Audio-Visual Speech Recognition

Training Transformer-based models demands a large amount of data, while obtaining aligned and labelled data in multimodality is rather cost-demanding, especially for audio-visual speech recognition (AVSR). Thus it makes a lot of sense to make use of unlabelled unimodal data. On the other side, although the effectiveness of large-scale self-supervised learning is well established in both audio and visual modalities, how to integrate those pre-trained models into a multimodal scenario remains underexplored. In this work, we successfully leverage unimodal self-supervised learning to promote the multimodal AVSR. In particular, audio and visual front-ends are trained on large-scale unimodal datasets, then we integrate components of both front-ends into a larger multimodal framework which learns to recognize parallel audio-visual data into characters through a combination of CTC and seq2seq decoding. We show that both components inherited from unimodal self-supervised learning cooperate well, resulting in that the multimodal framework yields competitive results through fine-tuning. Our model is experimentally validated on both word-level and sentence-level tasks. Especially, even without an external language model, our proposed model raises the state-of-the-art performances on the widely accepted Lip Reading Sentences 2 (LRS2) dataset by a large margin, with a relative improvement of 30%.

cs.SD↗

arXiv2022

DeepShovel: An Online Collaborative Platform for Data Extraction in Geoscience Literature with AI Assistance

Geoscientists, as well as researchers in many fields, need to read a huge amount of literature to locate, extract, and aggregate relevant results and data to enable future research or to build a scientific database, but there is no existing system to support this use case well. In this paper, based on the findings of a formative study about how geoscientists collaboratively annotate literature and extract and aggregate data, we proposed DeepShovel, a publicly-available AI-assisted data extraction system to support their needs. DeepShovel leverages the state-of-the-art neural network models to support researcher(s) easily and accurately annotate papers (in the PDF format) and extract data from tables, figures, maps, etc. in a human-AI collaboration manner. A follow-up user evaluation with 14 researchers suggested DeepShovel improved users' efficiency of data extraction for building scientific databases, and encouraged teams to form a larger scale but more tightly-coupled collaboration.

cs.HC↗

arXiv2022

On the Similarity between von Neumann Graph Entropy and Structural Information: Interpretation, Computation, and Applications

The von Neumann graph entropy is a measure of graph complexity based on the Laplacian spectrum. It has recently found applications in various learning tasks driven by networked data. However, it is computational demanding and hard to interpret using simple structural patterns. Due to the close relation between Lapalcian spectrum and degree sequence, we conjecture that the structural information, defined as the Shannon entropy of the normalized degree sequence, might be a good approximation of the von Neumann graph entropy that is both scalable and interpretable. In this work, we thereby study the difference between the structural information and von Neumann graph entropy named as {\em entropy gap}. Based on the knowledge that the degree sequence is majorized by the Laplacian spectrum, we for the first time prove the entropy gap is between $0$ and $\log_2 e$ in any undirected unweighted graphs. Consequently we certify that the structural information is a good approximation of the von Neumann graph entropy that achieves provable accuracy, scalability, and interpretability simultaneously. This approximation is further applied to two entropy-related tasks: network design and graph similarity measure, where novel graph similarity measure and fast algorithms are proposed. Our experimental results on graphs of various scales and types show that the very small entropy gap readily applies to a wide range of graphs and weighted graphs. As an approximation of the von Neumann graph entropy, the structural information is the only one that achieves both high efficiency and high accuracy among the prominent methods. It is at least two orders of magnitude faster than SLaQ with comparable accuracy. Our structural information based methods also exhibit superior performance in two entropy-related tasks.

cs.SI↗

arXiv2021

DeCOM: Decomposed Policy for Constrained Cooperative Multi-Agent Reinforcement Learning

In recent years, multi-agent reinforcement learning (MARL) has presented impressive performance in various applications. However, physical limitations, budget restrictions, and many other factors usually impose \textit{constraints} on a multi-agent system (MAS), which cannot be handled by traditional MARL frameworks. Specifically, this paper focuses on constrained MASes where agents work \textit{cooperatively} to maximize the expected team-average return under various constraints on expected team-average costs, and develops a \textit{constrained cooperative MARL} framework, named DeCOM, for such MASes. In particular, DeCOM decomposes the policy of each agent into two modules, which empowers information sharing among agents to achieve better cooperation. In addition, with such modularization, the training algorithm of DeCOM separates the original constrained optimization into an unconstrained optimization on reward and a constraints satisfaction problem on costs. DeCOM then iteratively solves these problems in a computationally efficient manner, which makes DeCOM highly scalable. We also provide theoretical guarantees on the convergence of DeCOM's policy update algorithm. Finally, we validate the effectiveness of DeCOM with various types of costs in both toy and large-scale (with 500 agents) environments.

cs.LG↗

arXiv2021

Scientific X-ray

The rapid development of modern science and technology has spawned rich scientific topics to research and endless production of literature in them. Just like X-ray imaging in medicine, can we intuitively identify the development limit and internal evolution pattern of scientific topic from the relationship of massive knowledge? To answer this question, we collect 71431 seminal articles of topics that cover 16 disciplines and their citation data, and extracts the "idea tree" of each topic to restore the structure of the development of 71431 topic networks from scratch. We define the Knowledge Entropy (KE) metric, and the contribution of high knowledge entropy nodes to increase the depth of the idea tree is regarded as the basis for topic development. By observing "X-ray images" of topics, We find two interesting phenomena: (1) Even though the scale of topics may increase unlimitedly, there is an insurmountable cap of topic development: the depth of the idea tree does not exceed 6 jumps, which coincides with the classical "Six Degrees of Separation"! (2) It is difficult for a single article to contribute more than 3 jumps to the depth of its topic, to this end, the continuing increase in the depth of the idea tree needs to be motivated by the influence relay of multiple high knowledge entropy nodes. Through substantial statistical fits, we derive a unified quantitative relationship between the change in topic depth ${ΔD}^t(v)$ and the change in knowledge entropy over time ${KE}^t\left(v\right)$ of the article $v$ driving the increase in depth in the topic: ${ΔD}^t(v) \approx \log \frac{KE^{t}(v)}{\left(t-t_{0}\right)^{1.8803}}$ , which can effectively portray evolution patterns of topics and predict their development potential. The various phenomena found by scientific x-ray may provide a new paradigm for explaining and understanding the evolution of science and technology.

cs.SI↗

arXiv2021

Influence of CO2-laser pulse parameters on 13.5 nm extreme ultraviolet emission features from irradiated liquid tin target

Laser-produced plasma (LPP) induced during irradiation of a liquid tin droplet with diameter of 150 um and 180 um by CO2 laser pulse with various pulse durations and energies is considered. The two-dimensional radiative magnetohydrodynamic (RMHD) plasma code is used to simulate the emission and plasma dynamics of multicharged ion tin LPP. Results of simulations for various laser pulse durations and 75-600 mJ pulse energies with Gaussian and experimentally taken temporal profiles are discussed. It is found that if the mass of the target is big enough to provide the plasma flux required (the considered case) a kind of dynamic quasi-stationary plasma flux is formed. In this dynamic quasi-stationary plasma flux, an interlayer of relatively cold tin vapor with mass density of 1-2 g/cm3 is formed between the liquid tin droplet and low density plasma of the critical layer. Expanding of the tin vapor from the droplet provides the plasma flux to the critical layer. In critical layer the plasma is heated up and expands faster. In the simulation results with spherical liquid tin target, the CE into 2$π$ is of 4% for 30 ns FWHM and just slightly lower - of 3.67% for 240 ns FWHM for equal laser intensities of 14 GW/cm2. This slight decay of the in-band EUV yield with laser pulse duration is conditioned by an increasing of radiation re-absorption by expanding plasma from the target, as more cold plasma is produced with longer pulse. The calculated direction diagrams of in-band EUV emission permit to optimize a collector configuration.

physics.plasm-ph↗

arXiv2021

Privacy Threats Analysis to Secure Federated Learning

Federated learning is emerging as a machine learning technique that trains a model across multiple decentralized parties. It is renowned for preserving privacy as the data never leaves the computational devices, and recent approaches further enhance its privacy by hiding messages transferred in encryption. However, we found that despite the efforts, federated learning remains privacy-threatening, due to its interactive nature across different parties. In this paper, we analyze the privacy threats in industrial-level federated learning frameworks with secure computation, and reveal such threats widely exist in typical machine learning models such as linear regression, logistic regression and decision tree. For the linear and logistic regression, we show through theoretical analysis that it is possible for the attacker to invert the entire private input of the victim, given very few information. For the decision tree model, we launch an attack to infer the range of victim's private inputs. All attacks are evaluated on popular federated learning frameworks and real-world datasets.

cs.LG↗

arXiv2021

Exploring the Disproportion Between Scientific Productivity and Knowledge Amount

The pursuit of knowledge is the permanent goal of human beings. Scientific literature, as the major medium that carries knowledge between scientists, exhibits explosive growth during the last century. Despite the frequent use of many tangible measures, such as citation, impact factor and g-index, to quantify the influence of papers from different perspectives based on scientific productivity, it has not yet been well understood how the relationship between scientific productivity and knowledge amount turns out to be, i.e., how the knowledge value of papers and knowledge amount vary with development of the discipline. This raises the question of whether high scientific productivity equals large knowledge amount. Here, building on rich literature on academic conferences and journals, we collect 185 million articles covering 19 disciplines published during 1970 to 2020, and establish citation network research area to represent the knowledge flow from the authors of the article being cited to the authors of the articles that cite it under each specific area. As a result, the structure formed during the evolution of each scientific area can implicitly tells how the knowledge flows between nodes and how it behaves as the number of literature (productivity) increases. By leveraging Structural entropy in structured high-dimensional space and Shannon entropy in unstructured probability space, we propose the Quantitative Index of Knowledge (KQI), which is taken as the subtraction between the two types of entropy, to reflect the extent of disorder difference (knowledge amount) caused by structure (order). With the aid of KQI, we find that, although the published literature shows an explosive growth, the amount of knowledge (KQI) contained in it obviously slows down, and there is a threshold after which the growth of knowledge accelerates...

cs.DL↗

arXiv2021

Certified Distributional Robustness on Smoothed Classifiers

The robustness of deep neural networks (DNNs) against adversarial example attacks has raised wide attention. For smoothed classifiers, we propose the worst-case adversarial loss over input distributions as a robustness certificate. Compared with previous certificates, our certificate better describes the empirical performance of the smoothed classifiers. By exploiting duality and the smoothness property, we provide an easy-to-compute upper bound as a surrogate for the certificate. We adopt a noisy adversarial learning procedure to minimize the surrogate loss to improve model robustness. We show that our training method provides a theoretically tighter bound over the distributional robust base classifiers. Experiments on a variety of datasets further demonstrate superior robustness performance of our method over the state-of-the-art certified or heuristic methods.

cs.LG↗

arXiv2021

Improved Matrix Gaussian Mechanism for Differential Privacy

The wide deployment of machine learning in recent years gives rise to a great demand for large-scale and high-dimensional data, for which the privacy raises serious concern. Differential privacy (DP) mechanisms are conventionally developed for scalar values, not for structural data like matrices. Our work proposes Improved Matrix Gaussian Mechanism (IMGM) for matrix-valued DP, based on the necessary and sufficient condition of $ (\varepsilon,δ) $-differential privacy. IMGM only imposes constraints on the singular values of the covariance matrices of the noise, which leaves room for design. Among the legitimate noise distributions for matrix-valued DP, we find the optimal one turns out to be i.i.d. Gaussian noise, and the DP constraint becomes a noise lower bound on each element. We further derive a tight composition method for IMGM. Apart from the theoretical analysis, experiments on a variety of models and datasets also verify that IMGM yields much higher utility than the state-of-the-art mechanisms at the same privacy guarantee.

cs.CR↗