arXiv subjects

Qi Su

Publications and source records attributed to Qi Su.

At least 55 records · Page 3Linked to original sources

arXiv2023

Incorporating Deep Syntactic and Semantic Knowledge for Chinese Sequence Labeling with GCN

Recently, it is quite common to integrate Chinese sequence labeling results to enhance syntactic and semantic parsing. However, little attention has been paid to the utility of hierarchy and structure information encoded in syntactic and semantic features for Chinese sequence labeling tasks. In this paper, we propose a novel framework to encode syntactic structure features and semantic information for Chinese sequence labeling tasks with graph convolutional networks (GCN). Experiments on five benchmark datasets, including Chinese word segmentation and part-of-speech tagging, demonstrate that our model can effectively improve the performance of Chinese labeling tasks.

cs.CL↗

arXiv2023

Prompt What You Need: Enhancing Segmentation in Rainy Scenes with Anchor-based Prompting

Semantic segmentation in rainy scenes is a challenging task due to the complex environment, class distribution imbalance, and limited annotated data. To address these challenges, we propose a novel framework that utilizes semi-supervised learning and pre-trained segmentation foundation model to achieve superior performance. Specifically, our framework leverages the semi-supervised model as the basis for generating raw semantic segmentation results, while also serving as a guiding force to prompt pre-trained foundation model to compensate for knowledge gaps with entropy-based anchors. In addition, to minimize the impact of irrelevant segmentation masks generated by the pre-trained foundation model, we also propose a mask filtering and fusion mechanism that optimizes raw semantic segmentation results based on the principle of minimum risk. The proposed framework achieves superior segmentation performance on the Rainy WCity dataset and is awarded the first prize in the sub-track of STRAIN in ICME 2023 Grand Challenges.

cs.CV↗

arXiv2023

The emergence of burstiness in temporal networks

Human social interactions tend to vary in intensity over time, whether they are in person or online. Variable rates of interaction in structured populations can be described by networks with the time-varying activity of links and nodes. One of the key statistics to summarize temporal patterns is the inter-event time (IET), namely the duration between successive pairwise interactions. Empirical studies have found IET distributions that are heavy-tailed (or "bursty"), for temporally varying interaction, both physical and digital. But it is difficult to construct theoretical models of time-varying activity on a network that reproduces the burstiness seen in empirical data. Here we develop a spanning-tree method to construct temporal networks and activity patterns with bursty behavior. Our method ensures a desired target IET distribution of single nodes/links, provided the distribution fulfills a consistency condition, regardless of whether the underlying topology is static or time-varying. We show that this model can reproduce burstiness found in empirical datasets, and so it may serve as a basis for studying dynamic processes in real-world bursty interactions.

physics.soc-ph↗

arXiv2022

Dim-Krum: Backdoor-Resistant Federated Learning for NLP with Dimension-wise Krum-Based Aggregation

Despite the potential of federated learning, it is known to be vulnerable to backdoor attacks. Many robust federated aggregation methods are proposed to reduce the potential backdoor risk. However, they are mainly validated in the CV field. In this paper, we find that NLP backdoors are hard to defend against than CV, and we provide a theoretical analysis that the malicious update detection error probabilities are determined by the relative backdoor strengths. NLP attacks tend to have small relative backdoor strengths, which may result in the failure of robust federated aggregation methods for NLP attacks. Inspired by the theoretical results, we can choose some dimensions with higher backdoor strengths to settle this issue. We propose a novel federated aggregation algorithm, Dim-Krum, for NLP tasks, and experimental results validate its effectiveness.

cs.LG↗

arXiv2022

GA-SAM: Gradient-Strength based Adaptive Sharpness-Aware Minimization for Improved Generalization

Recently, Sharpness-Aware Minimization (SAM) algorithm has shown state-of-the-art generalization abilities in vision tasks. It demonstrates that flat minima tend to imply better generalization abilities. However, it has some difficulty implying SAM to some natural language tasks, especially to models with drastic gradient changes, such as RNNs. In this work, we analyze the relation between the flatness of the local minimum and its generalization ability from a novel and straightforward theoretical perspective. We propose that the shift of the training and test distributions can be equivalently seen as a virtual parameter corruption or perturbation, which can explain why flat minima that are robust against parameter corruptions or perturbations have better generalization performances. On its basis, we propose a Gradient-Strength based Adaptive Sharpness-Aware Minimization (GA-SAM) algorithm to help to learn algorithms find flat minima that generalize better. Results in various language benchmarks validate the effectiveness of the proposed GA-SAM algorithm on natural language tasks.

cs.LG↗

arXiv2022

Automatic Translating between Ancient Chinese and Contemporary Chinese with Limited Aligned Corpora

The Chinese language has evolved a lot during the long-term development. Therefore, native speakers now have trouble in reading sentences written in ancient Chinese. In this paper, we propose to build an end-to-end neural model to automatically translate between ancient and contemporary Chinese. However, the existing ancient-contemporary Chinese parallel corpora are not aligned at the sentence level and sentence-aligned corpora are limited, which makes it difficult to train the model. To build the sentence level parallel training data for the model, we propose an unsupervised algorithm that constructs sentence-aligned ancient-contemporary pairs by using the fact that the aligned sentence pair shares many of the tokens. Based on the aligned corpus, we propose an end-to-end neural model with copying mechanism and local attention to translate between ancient and contemporary Chinese. Experiments show that the proposed unsupervised algorithm achieves 99.4% F1 score for sentence alignment, and the translation model achieves 26.95 BLEU from ancient to contemporary, and 36.34 BLEU from contemporary to ancient.

cs.CL↗

arXiv2022

That Slepen Al the Nyght with Open Ye! Cross-era Sequence Segmentation with Switch-memory

The evolution of language follows the rule of gradual change. Grammar, vocabulary, and lexical semantic shifts take place over time, resulting in a diachronic linguistic gap. As such, a considerable amount of texts are written in languages of different eras, which creates obstacles for natural language processing tasks, such as word segmentation and machine translation. Although the Chinese language has a long history, previous Chinese natural language processing research has primarily focused on tasks within a specific era. Therefore, we propose a cross-era learning framework for Chinese word segmentation (CWS), CROSSWISE, which uses the Switch-memory (SM) module to incorporate era-specific linguistic knowledge. Experiments on four corpora from different eras show that the performance of each corpus significantly improves. Further analyses also demonstrate that the SM can effectively integrate the knowledge of the eras into the neural network.

cs.CL↗

arXiv2022

The arrow of evolution when the offspring variance is large

The concept of fitness is central to evolution, but it quantifies only the expected number of offspring an individual will produce. The actual number of offspring is also subject to noise, arising from environmental or demographic stochasticity. In nature, individuals who are more fecund tend to have greater variance in their offspring number -- sometimes far greater than the Poisson variance assumed in classical models of population genetics. Here, we develop a model for the evolution of two types reproducing in a population of non-constant size. The frequency-dependent fitness of each type is determined by pairwise interactions in a prisoner's dilemma game, but the offspring number is subject to an exogenously controlled variance that may depend upon the mean. Whereas defectors are preferred by natural selection in classical well-mixed populations, since they always have greater fitness than cooperators, we show that large offspring variance can reverse the direction of evolution and favor cooperation. Reproductive over-dispersion produces qualitatively new dynamics for other types of social interactions, as well, which cannot arise in populations with a fixed size or Poisson offspring variance.

q-bio.PE↗

arXiv2022

Adversarial Parameter Defense by Multi-Step Risk Minimization

Previous studies demonstrate DNNs' vulnerability to adversarial examples and adversarial training can establish a defense to adversarial examples. In addition, recent studies show that deep neural networks also exhibit vulnerability to parameter corruptions. The vulnerability of model parameters is of crucial value to the study of model robustness and generalization. In this work, we introduce the concept of parameter corruption and propose to leverage the loss change indicators for measuring the flatness of the loss basin and the parameter robustness of neural network parameters. On such basis, we analyze parameter corruptions and propose the multi-step adversarial corruption algorithm. To enhance neural networks, we propose the adversarial parameter defense algorithm that minimizes the average risk of multiple adversarial parameter corruptions. Experimental results show that the proposed algorithm can improve both the parameter robustness and accuracy of neural networks.

cs.LG↗

arXiv2022

Chinese Word Segmentation with Heterogeneous Graph Neural Network

In recent years, deep learning has achieved significant success in the Chinese word segmentation (CWS) task. Most of these methods improve the performance of CWS by leveraging external information, e.g., words, sub-words, syntax. However, existing approaches fail to effectively integrate the multi-level linguistic information and also ignore the structural feature of the external information. Therefore, in this paper, we proposed a framework to improve CWS, named HGNSeg. It exploits multi-level external information sufficiently with the pre-trained language model and heterogeneous graph neural network. The experimental results on six benchmark datasets (e.g., Bakeoff 2005, Bakeoff 2008) validate that our approach can effectively improve the performance of Chinese word segmentation. Importantly, in cross-domain scenarios, our method also shows a strong ability to alleviate the out-of-vocabulary (OOV) problem.

cs.CL↗

arXiv2021

One-shot Weakly-Supervised Segmentation in Medical Images

Deep neural networks usually require accurate and a large number of annotations to achieve outstanding performance in medical image segmentation. One-shot segmentation and weakly-supervised learning are promising research directions that lower labeling effort by learning a new class from only one annotated image and utilizing coarse labels instead, respectively. Previous works usually fail to leverage the anatomical structure and suffer from class imbalance and low contrast problems. Hence, we present an innovative framework for 3D medical image segmentation with one-shot and weakly-supervised settings. Firstly a propagation-reconstruction network is proposed to project scribbles from annotated volume to unlabeled 3D images based on the assumption that anatomical patterns in different human bodies are similar. Then a dual-level feature denoising module is designed to refine the scribbles based on anatomical- and pixel-level features. After expanding the scribbles to pseudo masks, we could train a segmentation model for the new class with the noisy label training strategy. Experiments on one abdomen and one head-and-neck CT dataset show the proposed method obtains significant improvement over the state-of-the-art methods and performs robustly even under severe class imbalance and low contrast.

eess.IV↗

arXiv2021

Evolution of prosocial behavior in multilayer populations

Human societies include diverse social relationships. Friends, family, business colleagues, and online contacts can all contribute to one's social life. Individuals may behave differently in different domains, but success in one domain may engender success in another. Here, we study this problem using multilayer networks to model multiple domains of social interactions, in which individuals experience different environments and may express different behaviors. We provide a mathematical analysis and find that coupling between layers tends to promote prosocial behavior. Even if prosociality is disfavored in each layer alone, multilayer coupling can promote its proliferation in all layers simultaneously. We apply this analysis to six real-world multilayer networks, ranging from the socio-emotional and professional relationships in a Zambian community, to the online and offline relationships within an academic University. We discuss the implications of our results, which suggest that small modifications to interactions in one domain may catalyze prosociality in a different domain.

physics.soc-ph↗

arXiv2021

Evolution of cooperation with joint liability

"Personal responsibility", one of the basic principles of modern law, requires one to be responsible for what he did. However, personal responsibility is far from the only norm ruling human interactions, especially in social and economic activities. In many collective communities such as among enterprise colleagues and family members, one's personal interests are often bound to others' -- once one member breaks the rule, a group of people have to bear the punishment or sanction. Such a mechanism is termed "joint liability". Although many real-world cases have demonstrated that joint liability helps to maintain collective collaboration, a deep and systematic theoretical analysis on how and when joint liability promotes cooperation is lacking. Here we use evolutionary game theory to model an interacting system with joint liability, where one's losing credit could deteriorate the reputation of the whole group. We provide the analytical condition to predict when cooperation evolves in the presence of joint liability, which is verified by simulations. We also analytically prove that joint liability can greatly promote cooperation. Our work stresses that joint liability is of great significance in promoting the current economic propensity.

physics.soc-ph↗

arXiv2021

CorbFuzz: Checking Browser Security Policies with Fuzzing

Browsers use security policies to block malicious behaviors. Cross-Origin Read Blocking (CORB) is a browser security policy for preventing side-channel attacks such as Spectre. We propose a web browser security policy fuzzer called CorbFuzz for checking CORB and similar policies. In implementing a security policy, the browser only has access to HTTP requests and responses, and takes policy actions based solely on those interactions. In checking the browser security policies, CorbFuzz uses a policy oracle that tracks the web application behavior and infers the desired policy action based on the web application state. By comparing the policy oracle with the browser behavior, CorbFuzz detects weaknesses in browser security policies. CorbFuzz checks the web browser policy by fuzzing a set of web applications where the state-related queries are symbolically evaluated for increased coverage and automation. CorbFuzz collects type information from database queries and branch conditions in order to prevent the generation of inconsistent data values during fuzzing. We evaluated CorbFuzz on CORB implementations of Chromium and Webkit, and Opaque Response Blocking (ORB) policy implementation of Firefox using web applications collected from GitHub. We found three classes of weaknesses in Chromium's implementation of CORB.

cs.CR↗

arXiv2021

Payoff Control in Repeated Games

Evolutionary game theory is a powerful mathematical framework to study how intelligent individuals adjust their strategies in collective interactions. It has been widely believed that it is impossible to unilaterally control players' payoffs in games, since payoffs are jointly determined by all players. Until recently, a class of so-called zero-determinant strategies are revealed, which enables a player to make a unilateral payoff control over her partners in two-action repeated games with a constant continuation probability. The existing methods, however, lead to the curse of dimensionality when the complexity of games increases. In this paper, we propose a new mathematical framework to study ruling strategies (with which a player unilaterally makes a linear relation rule on players' payoffs) in repeated games with an arbitrary number of actions or players, and arbitrary continuation probability. We establish an existence theorem of ruling strategies and develop an algorithm to find them. In particular, we prove that strict Markov ruling strategy exists only if either the repeated game proceeds for an infinite number of rounds, or every round is repeated with the same probability. The proposed mathematical framework also enables the search of collaborative ruling strategies for an alliance to control outsiders. Our method provides novel theoretical insights into payoff control in complex repeated games, which overcomes the curse of dimensionality.

math.OC↗

arXiv2021

Alleviating the Knowledge-Language Inconsistency: A Study for Deep Commonsense Knowledge

Knowledge facts are typically represented by relational triples, while we observe that some commonsense facts are represented by the triples whose forms are inconsistent with the expression of language. This inconsistency puts forward a challenge for pre-trained language models to deal with these commonsense knowledge facts. In this paper, we term such knowledge as deep commonsense knowledge and conduct extensive exploratory experiments on it. We show that deep commonsense knowledge occupies a significant part of commonsense knowledge while conventional methods fail to capture it effectively. We further propose a novel method to mine the deep commonsense knowledge distributed in sentences, alleviating the reliance of conventional methods on the triple representation form of knowledge. Experiments demonstrate that the proposal significantly improves the performance in mining deep commonsense knowledge.

cs.CL↗

arXiv2021

Evolution of cooperation with asymmetric social interactions

How cooperation emerges in human societies is both an evolutionary enigma, and a practical problem with tangible implications for societal health. Population structure has long been recognized as a catalyst for cooperation because local interactions enable reciprocity. Analysis of this phenomenon typically assumes bi-directional social interactions, even though real-world interactions are often uni-directional. Uni-directional interactions -- where one individual has the opportunity to contribute altruistically to another, but not conversely -- arise in real-world populations as the result of organizational hierarchies, social stratification, popularity effects, and endogenous mechanisms of network growth. Here we expand the theory of cooperation in structured populations to account for both uni- and bi-directional social interactions. Even though directed interactions remove the opportunity for reciprocity, we find that cooperation can nonetheless be favored in directed social networks and that cooperation is provably maximized for networks with an intermediate proportion of directed interactions, as observed in many empirical settings. We also identify two simple structural motifs that allow efficient modification of interaction directionality to promote cooperation by orders of magnitude. We discuss how our results relate to the concepts of generalized and indirect reciprocity.

q-bio.PE↗

arXiv2020

Memorized Sparse Backpropagation

Neural network learning is usually time-consuming since backpropagation needs to compute full gradients and backpropagate them across multiple layers. Despite its success of existing works in accelerating propagation through sparseness, the relevant theoretical characteristics remain under-researched and empirical studies found that they suffer from the loss of information contained in unpropagated gradients. To tackle these problems, this paper presents a unified sparse backpropagation framework and provides a detailed analysis of its theoretical characteristics. Analysis reveals that when applied to a multilayer perceptron, our framework essentially performs gradient descent using an estimated gradient similar enough to the true gradient, resulting in convergence in probability under certain conditions. Furthermore, a simple yet effective algorithm named memorized sparse backpropagation (MSBP) is proposed to remedy the problem of information loss by storing unpropagated gradients in memory for learning in the next steps. Experimental results demonstrate that the proposed MSBP is effective to alleviate the information loss in traditional sparse backpropagation while achieving comparable acceleration.

cs.LG↗