2021年第20期Seminar——李鹏程、郑克鲜

2021年第20期Seminar——李鹏程、郑克鲜

Talk 1

Title: The Hidden Geometry of Complex, Network-Driven Contagion Phenomena

Speaker: Pengcheng Li

Abstract:

The global spread of epidemics, rumors, opinions, and innovations are complex, network-driven dynamic processes. The combined multiscale nature and intrinsic heterogeneity of the underlying networks make it difficult to develop an intuitive understanding of these processes, to distinguish relevant from peripheral factors, to predict their time course, and to locate their origin. However, we show that complex spatiotemporal patterns can be reduced to surprisingly simple, homogeneous wave propagation patterns, if conventional geographic distance is replaced by a probabilistically motivated effective distance. In the context of global, air-traffic–mediated epidemics, we show that effective distance reliably predicts disease arrival times. Even if epidemiological parameters are unknown, the method can still deliver relative arrival times. The approach can also identify the spatial origin of spreading processes and successfully be applied to data of the worldwide 2009 H1N1 influenza pandemic and 2003 SARS epidemic.

Supervisor: Ze Kang

Talk 2

Title: Percolation of interdependent networks with conditional dependency clusters

Speaker: Kexian Zheng

Abstract:

Modern systems are always coupled. Previous studies indicate that coupled systems are more fragile than single systems. In a single system, when a fraction of 1-p nodes are removed, the percolation process is often of the second order. In a coupled system, due to the lack of resilience, the phase transition is always of the first order when removing a fraction of nodes. Most of previous studies on coupled systems focus on one-to-one dependency relation. This kind of relationship is called a no-feedback condition. Existing studies suppose that coupled systems are much more fragile without a no-feedback condition. That is to say, if a node depends on more than one node, the coupled system will breakdown even when a small fraction of nodes are removed from the coupled system. By observing the real world system, real nodes are often dependent on a dependency cluster, which consists of more than one other node. For example, in an industry chain, an electronic equipment factory may need several raw material factories to supply production components. Despite part of the raw material factories being bankrupt, the electronic equipment factory can carry out production normally because the remaining raw material factories still supply the necessary production components. But theoretical analysis shows that the robustness of such a coupled system is worse than that of one-to-one dependency system. Actually, the coupled system in real world does not usually disintegrate into pieces after some nodes have become invalid. To explain this phenomenon, we model a coupled system as interdependent networks and study, both analytically and numerically, the percolation in interdependent networks with conditional dependency clusters. A node in our model survives until the number of failed nodes in its dependency cluster is greater than a threshold. Our exact solutions of giant component size are in good agreement with the simulation results. Though our model does not have second order phase transition, we still find ways to improve the robustness of interdependent networks. One way is to increase the dependency cluster failure threshold. A higher threshold means that more nodes in the dependency cluster can be removed without breaking down the node depending on the cluster. Other way is to increase the size of dependency clusters, the more the nodes in the dependency cluster, the more the failure combinations are, which increases the survival probability of the node depending on cluster. Our model offers a useful strategy to enhance the robustness of coupled system and makes a good contribution to the study of interdependent networks with dependency clusters.

Supervisor: Jianxin Pei

 

Time:16:00  April 15, 2021

Address:MingLi Buliding C1102

Chair: Bo Miao

 

2021年第19期Seminar——康泽、阳成

2021年第19期Seminar——康泽、阳成

Talk 1

Title: Anomalous role of information diffusion in epidemic spreading

Speaker: Ze Kang

Abstract:

Approximate Bayesian computation (ABC) constitutes a class of computational methods rooted in Bayesian statistics. In all model-based statistical inference, the likelihood function is of central importance, since it expresses the probability of the observed data under a particular statistical model, and thus quantifies the support data lend to particular values of parameters and to choices among different models. For simple models, an analytical formula for the likelihood function can typically be derived. However, for more complex models, an analytical formula might be elusive or the likelihood function might be computationally very costly to evalu- ate. ABC methods bypass the evaluation of the likelihood function. In this way, ABC methods widen the realm of models for which statistical inference can be considered. ABC methods are mathematically well-founded, but they inevitably make assumptions and approximations whose impact needs to be carefully assessed. Furthermore, the wider application domain of ABC exacerbates the challenges of parameter estimation and model selection. ABC has rapidly gained popularity over the last years and in particular for the analysis of complex problems arising in biological sciences (e.g., in population genetics, ecology, epidemiology, and systems biology).

Supervisor: Lifeng Shen

Talk 2

Title: Information Source Detection in the SIR Model

Speaker: Cheng Yang

Abstract:

This paper studies the problem of detecting the information source in a network in which the spread of information follows the popular Susceptible-Infected-Recovered (SIR) model. We assume all nodes in the network are in the susceptible state initially except the information source which is in the infected state. Susceptible nodes may then be infected by infected nodes, and infected nodes may recover and will not be infected again after recovery. Given a snapshot of the network, from which we know all infected nodes but cannot distinguish susceptible nodes and recovered nodes, the problem is to find the information source based on the snapshot and the network topology. We develop a sample path based approach where the estimator of the information source is chosen to be the root node associated with the sample path that most likely leads to the observed snapshot. We prove for infinite-trees, the estimator is a node that minimizes the maximum distance to the infected nodes. A reverse-infection algorithm is proposed to find such an estimator in general graphs. We prove that for g-regular trees such that gq > 1, where g is the node degree and q is the infection probability, the estimator is within a constant distance from the actual source with a high probability, independent of the number of infected nodes and the time the snapshot is taken. Our simulation results show that for tree networks, the estimator produced by the reverse-infection algorithm is closer to the actual source than the one identified by the closeness centrality heuristic. We then further evaluate the performance of the reverse infection algorithm on several real world networks.

Supervisor: PengCheng Li

 

Time:16:00  April 8, 2021

Address:MingLi Buliding C1102

Chair: PengCheng Li

 

 

2021年第18期Seminar——郭志成、李丽

2021年第18期Seminar——郭志成、李丽

Talk 1

Title: Anomalous role of information diffusion in epidemic spreading

Speaker: Zhicheng Guo

Abstract:

A widely held belief in network epidemiology is that information diffusion making the individuals aware of the epidemic and thus driving them to seek protections from non-pharmaceutical or pharmaceutical resources can help suppress its spreading. However, as the COVID-19 pandemic has revealed, excessive information diffusion can trigger irrational acquisition and hoarding behaviors, which can lead to shortages of the resources even for those in urgent need and consequently, worsening the disease spreading. To develop a quantitative understanding of the effect of information diffusion on epidemic spreading subject to allocations of limited resources has become an urgently important problem with broad implications. We construct a multiplex network framework to characterize the complex interplay among resource allocation, information diffusion, and epidemic spreading, and develop a microscopic Markov chain theory to analyze their coevolution dynamics. There are two mainndings. Firstly, if the infected individuals have a large recovery probability, information diffusion plays the expected \normal” role of suppressing the epidemic. However, if the recovery probability is low, information diffusion can anomalously worsen the spreading, regardless of the available resources insofar as they are limited. Secondly, different types of resources can lead to distinct phase transitions underlying the epidemic outbreak: with abundant cure focused resources, the phase transition is of the second order, but if the resources are of the protection type and they are not as abundant, the transition becomes first order and a hysteresis loop emerges. The generality of the findings is established through simulations of synthetic and empirical three-layer networks with results in agreement with the theoretical predictions.

Supervisor: Kexian Zheng

Talk 2

Title: Aspect-Level Sentiment Analysis Via Convolution over Dependency Tree

Speaker: Li Li

Abstract:

We propose a method based on neural networks to identify the sentiment polarity of opinion words expressed on a specific aspect of a sentence. Although a large majority of works typically focus on leveraging the expressive power of neural networks in handling this task, we explore the possibility of integrating dependency trees with neural networks for representation learning. To this end, we present a convolution over a dependency tree (CDT) model which exploits a Bi-directional Long Short Term Memory (Bi-LSTM) to learn representations for features of a sentence, and further enhance the embeddings with a graph convolutional network (GCN) which operates directly on the dependency tree of the sentence. Our approach propagates both contextual and dependency information from opinion words to aspect words, offering discriminative properties for supervision. Experimental results ranks our approach as the new state of-the-art in aspect-based sentiment classification.

Supervisor: Wei Zhang

 

Time:16:00  April 1, 2021

Address:MingLi Buliding C1102

Chair: Ze Kang

2021年第17期Seminar报告总结

2021年第17期Seminar报告总结

第一位主讲人是20级的黄玉楠,她此次为大家分享的是一篇名为《Graph Neural Network-Based Anomaly Detection in Multivariate Time Series》的论文。

主要是从相关工作、模型架构和实验展开分享。首先是相关工作部分,作者在文中提到传统的异常检测模型只能对数据进行一个简单的建模,比如线性拟合,不能处理更加复杂的非线性关系的数据。随着深度学习技术的发展,基于深度学习技术的异常检测模型在处理复杂的高维数据方面取得不错的效果,但这些方法虽然能处理高维数据复杂数据,并没有去专门地学习各特征之间的内在依赖关系,这就导致这些模型在对具有大量潜在依赖关系的复杂数据建模时的局限性。而基于图的方法的提出,为对特征依赖关系建模提供了一个思路,图神经网络在近年来已经被成功地应用在对复杂的图结构数据建模上,所以,作者基于上述工作提出了一种基于图神经网络的异常检测模型。

模型分别先对时序数据进行embedding操作,再根据embedding向量学习各传感器之间的依赖关系,然后对下一个时间戳进行预测,最后根据预测数据和观察数据进行偏移得分计算,根据阈值得出最后检测结果。最后对实验结果进行分析总结。

第二个出场的是20级的陈嘉豪,他此次为大家分享的是阴影绘制。

绘制正确的阴影对于生成一张具有真实感的图片非常重要,因此在渲染中对于阴影的绘制一直是一个非常重要的课题。在实时渲染中,由于光栅化图形管线缺少全局性息,与光线追踪不同,对于每一个着色点,我们无法知道它的全局光照信息,所以也无法知道该点是否位于阴影中。因此,我们在生成一张图片前需要一个预渲染阶段来判定着色点是否位于阴影中,而解决这一问题的方法被称为shadow mapping。在预渲染阶段,基于z-buffer算法从光源看向场景,记录每个像素对应的最小深度,生成一张shadow mapping,在正式渲染阶段,根据着色点到光源的距离与深度图记录的距离来判断该点是否在阴影中,如果两个距离相等,则说明不在阴影中,如果着色点到光源距离大于深度图记录的深度,说明着色点和光源之间有遮挡物,该点位于阴影中。Shadow mapping技术在今天任然被广泛采用,在所有3D电子游戏中,都使用的是shadow mapping技术,早年的3D动画也采用的是shadow mapping技术。直至今日,shadow mapping技术任然在向前发展,对于这一技术的改进一直层出不穷

以下是黄玉楠、陈嘉豪同学在这次seminar中的表现评分。

 

 

2021年第17期Seminar——黄玉楠、陈嘉豪

2021年第17期Seminar——黄玉楠、陈嘉豪

Talk 1

Title: Graph Neural Network-Based Anomaly Detection in Multivariate Time Series

Speaker: Yunan Huang

Abstract:

Given high-dimensional time series data (e.g., sensor data), how can we detect anomalous events, such as system faults and attacks? More challengingly, how can we do this in a way that captures complex inter-sensor relationships, and detects and explains anomalies which deviate from these relationships? Recently, deep learning approaches have enabled improvements in anomaly detection in high-dimensional datasets; however, existing methods do not explicitly learn the structure of existing relationships between variables, or use them to predict the expected behavior of time series. Our approach combines a structure learning approach with graph neural networks, additionally using attention weights to provide explainability for the detected anomalies. Experiments on two real-world sensor datasets with ground truth anomalies show that our method detects anomalies more accurately than baseline approaches, accurately captures correlations between sensors, and allows users to deduce the root cause of a detected anomaly.

Supervisor: Yongfang Dai

Talk 2

Title: Real-Time Rendering

Speaker: Jiahao Chen

Abstract:

Shadows are important for creating realistic images and in providing the user with visual cues about object placement. Because shadows are so important for realistic rendering, this area is getting a lot of attention, especially in real time, where we not only have to think about how to get a more realistic scene, but also have to be limited by hardware performance. These two goals — realism and performance — are difficult to reconcile completely, especially with shadow algorithms. Therefore, a large number of shadow algorithms have been proposed in recent years. In particular, the algorithm of soft shadow correlation has increasingly become a major research direction because it can obtain more physically reliable results, and the related works are frequently published.

Supervisor: Xiao Liang

 

Time:16:00  March 25, 2021

Address:MingLi Buliding C1102

Chair: Zhicheng Guo