2020年第8期Seminar——李鹏程、苗波

2020年第8期Seminar——李鹏程、苗波

Talk 1

Title:GRAPH-BERT :Characterizing the dynamics underlying global

Speaker: Pengcheng Li

Abstract:

Abstract—Over the past few decades, global metapopulation epidemic simulations built with worldwide air-transportation data have been the main tool for studying how epidemics spread from the origin to other parts of the world (e.g., for pandemic influenza, SARS, and Ebola). However, it remains unclear how disease epidemiology and the air-transportation network structure determine epidemic arrivals for different populations around the globe. Here, we fill this knowledge gap by developing and validating an analytical framework that requires only basic analytics from stochastic processes. We apply this framework retrospectively to the 2009 influenza pandemic and 2014 Ebola epidemic to show that key epidemic parameters could be robustly estimated in real-time from public data on local and global spread at very low computational cost. Our framework not only elucidates the dynamics underlying global spread of epidemics but also advances our capability in nowcasting and forecasting epidemics.

Supervisor: Ze Kang, Lifeng Shen

Talk 2

Title:Extracting Statistical Graph Features for Accurate and Efficient Time Series Classification

Speaker: Bo Miao

Abstract:

This paper presents a multiscale visibility graph representation for time series as well as feature extraction methods for time series classification (TSC). Unlike traditional TSC approaches that seek to find global similarities in time series databases (e.g., Nearest Neighbor with Dynamic Time Warping distance) or methods specializing in locating local patterns/subsequences (e.g.,shapelets), we extract solely statistical features from graphs that are generated from time series. Specifically, we augment time series by means of their multiscale approximations, which are further transformed into a set of visibility graphs. After extracting probability distributions of small motifs, density, assortativity, etc., these features are used for building highly accurate classification models using generic classifiers (e.g., Support Vector Machine and eXtreme Gradient Boosting). Thanks to the way how we transform time series into graphs and extract features from them, we are able to capture both global and local features from time series. Based on extensive experiments on a large number of open datasets and comparison with five state-of-the-art TSC algorithms, our approach is shown to be both accurate and efficient: it is more accurate than Learning Shapelets and at the same time faster than Fast Shapelets.

Supervisor: Yan Chen, Yongfang Dai

 

Time:16:00  October 29, 2020

Address:MingLi Buliding C1102

Chair: Na Chen

2020年第7期Seminar报告总结

2020年第7期Seminar报告总结

第一位主讲人是19级的邓丹,她此次为大家分享的是一篇名为《Deep Forest》的论文

邓丹同学首先介绍了DNN的三个不足之处,然后引出Deep Forest的概念,然后介绍了Gcforest模型示意图,对Gcforest的两大模块Multi-Grained ScanningCascade Forest做了详细的介绍。然后对决策树做了一个简单的介绍。介绍完两大模块之后,接着又介绍了Gcforest的流程图,从Multi-Grained ScanningCascade Forest的流程。

Gcforest流程图

接着,邓丹同学通过四个实验,将gcForest与深度神经网络和其他几种流行的学习算法进行比较。

验证了gcForest的超参数设置比深度神经网络容易得多,只对所有任务使用相同的设置。而针对DNNs执行特定任务的调优,分别从手写识别、人脸识别、音乐片段识别 ,手势识别等等方面的数据集进行验证,取得了和DNN相当甚至更好的结果

四个实验

最后,做了一个对Gcforest的总结

    第二个出场的是19级的张巍,他此次为大家分享的是一篇名为《GRAPH-BERT》的论文,本次讲解从图神经网络中存在的假死现象以及过平滑的问题开始,作者提出了GRAPH-BERT, 这种方法不需要依赖卷积、聚合的操作就可以实现图表示学习主要的思路是将原始图分解成以每一个节点为中心的多个子图,只利用attention机制在子图上进行表征学习,然后利用attention去学习结点表征,而不考虑子图中的边信息;另一方面也解决了大规模图的效率问题。

张巍同学在介绍GRAPH-BERT的方法

以下是邓丹、张巍同学在这次seminar中的表现评分。

 

 

2020年第7期Seminar——张巍、邓丹

2020年第7期Seminar——张巍、邓丹

Talk 1

Title:GRAPH-BERT : Only Attention is Needed for Learning Graph Representa-tions

Speaker: Wei Zhang

Abstract:

Abstract—The dominant graph neural networks (GNNs) over-rely on the graph links, several serious performance problems with which have been witnessed already, e.g., suspended animation problem and over-smoothing problem. What’s more, the inherently inter-connected nature precludes parallelization within the graph, which becomes critical for large-sized graph, as memory constraints limit batching across the nodes. In this paper, we will introduce a new graph neural network, namely G RAPH -B ERT (Graph based B ERT ), solely based on the attention mechanism without any graph convolution or aggregation operators. Instead of feeding G RAPH -B ERT with the complete large input graph, weproposetotrain G RAPH -B ERT withsampled linkless subgraphs within their local contexts.G RAPH -B ERT can be learned effectively in a standalone mode. Meanwhile, a pre-trained G RAPH -B ERT can also be transferred to other application tasks directly or with necessary fine-tuning if any supervised label information or certain application oriented objective is available. We have tested the effectiveness of G RAPH -B ERT on several graph benchmark datasets. Based the pretrained G RAPH -B ERT with the node attribute reconstruction and structure recovery tasks, we further fine-tune G RAPH -B ERT on node classification and graph clustering tasks specifically. The experimental results have demonstrated that G RAPH -B ERT can out-perform the existing GNNs in both the learning effectiveness and efficiency.

Supervisor: Yang Yang, Biao Wang

Talk 2

Title: Deep Forest

Speaker: Dan Deng

Abstract:

Current deep learning models are mostly build upon neural networks, i.e., multiple layers of parameterized differentiable nonlinear modules that can be trained by backpropagation. In this paper, we explore the possibility of building deep models based on non-differentiable modules. We conjecture that the mystery behind the success of deep neural networks owes much to three characteristics, i.e., layer-by-layer processing, in-model feature transformation and sucient model complexity. We propose the gcForest approach, which generates deep forest holding these characteristics. This is a decision tree ensemble approach, with much less hyper-parameters than deep neural networks, and its model complexity can be automatically determined in a data-dependent way. Experiments show that its performance is quite robust to hyper-parameter settings, such that in most cases, even across different data from different domains, it is able to get excellent performance by using the same default setting. This study opens the door of deep learning based on non-dierentiable modules, and exhibits the possibility of constructing deep models without using backpropagation.

Supervisor: Yuan Zhong

 

Time:16:00  October 22, 2020

Address:MingLi Buliding C1102

Chair: Pengcheng Li

2020年第6期Seminar报告总结

2020年第6期Seminar报告总结

第一位主讲人是19级的欧阳娇,她此次为大家分享的是一篇名为《Guided Image Filtering》的论文,这篇论文主要介绍了基于高斯滤波提出了双边滤波和引导滤波,并讲解了这两种滤波相比于高斯滤波的改进之处。

欧阳娇同学首先介绍了高斯滤波,高斯滤波就是对整幅图像进行加权平均的过程,每一个像素点的值,都由其本身和邻域内的其他像素值经过加权平均后得到,也指出高斯滤波的缺点是只考虑空间距离的权重,会对图像边缘造成模糊。

接着,欧阳同学又介绍了双边滤波,指出双边滤波在高斯滤波的基础上,进一步优化,叠加了像素值的考虑。在图像的低频区域,像素值变化很小,那么像素差值接近于0,对应的值域权重接近于1,此时空间域权重起主要在图像的边缘区域,像素值变化很大,即使距离近,空间域权重大。加上值域权重总的系数也较小,从而保护了边缘的信息。可见,双边滤波考虑了像素在空间和像素值的相似性,通过在高斯滤波的基础上增加一个衡量像素变化的滤波器,避免在滤波期间对边缘平滑作用,相当于进行高斯模糊

最后,欧阳同学介绍了引导滤波,引导滤波的作用就在于搜索出线性因数的最优解,使输入图像与输出图像之间的差值最小化,导向滤波不仅可以做到边缘保护过滤,非迭代,还具有快速的线性时间算法,目前,它是最快的边缘保持滤波器之一,而不管滤波模板的大小,还具有梯度保护。除此之外还可以将导向图像的结构转换为滤波输出,从而实现新的滤波应用,如去雾和导向羽化。总之,引导滤波在各种计算机视觉和计算机图形学应用中很多,包括边缘感知平滑、去噪、细节增强、HDR压缩、图像抠图/羽化、去雾、联合上采样等。

第二个出场的是19级的杨洋,她此次为大家分享的是一篇名为《Adversarial Attack on Community by Hiding Individuals》的论文,这篇论文主要讨论的是如何在社团检测中去隐藏一些目标或信息来实现隐私保护。

杨洋同学在做GNN对抗攻击的背景介绍

杨洋同学首先为大家介绍了涉及这篇论文的一些基础知识。从GNN对抗攻击的背景到这篇论文的中心思想,详细讲解了这篇论文的总体思想与设计思路。作者主要是利用受约束图生成器来获得受扰动之后的图,然后利用鲁棒性训练的思想将干净的图数据与受攻击后的图数据一起送入到代理社团检测模型中计算得到Lu。论文设计的模型框架如下:

在介绍了总体思想之后,杨洋同学开始介绍论文的总体设计框架。其中,constrained graph generation是基于VAE做的一些改进:encoder沿用VAE的,但是在generation部分进行了修改,因为有扰动的budget的限制和计算的伸缩性两个问题需要考虑,因此在考虑第一个问题的时候在budget部分使用了mask机制,能够阻止生成不需要的连边;第二个问题是对不同规模的图大小使用不同的解决方案(大图,删除边;小图,增、删边)。

最后杨洋根据实验数据对论文所提出的想法做出客观的评价,对论文中提出的思想做了总结。

以下是杨洋、欧阳娇同学在这次seminar中的表现评分。

 

2020年第6期Seminar——杨洋、欧阳娇

2020年第6期Seminar——杨洋、欧阳娇

Talk 1

Title:Adversarial Attack on Community Detection by Hiding Individuals

Speaker: Yang Yang

Abstract:

In this paper, we extend adversarial graphs to the problem of community detection which is much more difficult. We focus on black-box attack and aim to hide targeted individuals from the detection of deep graph community detection models, which has many applications in real-world scenarios, for example, protecting personal privacy in social networks and understanding camouflage patterns in transaction networks.We propose an iterative learning framework that takes turns to update two modules: one working as the constrained graph generator and the other as the surrogate community detection model. We also find that the adversarial graphs generated by our method can be transferred to other learning based community detection models.

Supervisor: Wei Zhang, Biao Wang

Talk 2

Title:Guided Image Filtering

Speaker: Yangjiao Ou

Abstract:

Abstract—In this paper, we propose a novel explicit image filter called guided filter. Derived from a local linear model, the guided filter computes the filtering output by considering the content of a guidance image, which can be the input image itself or another different image. The guided filter can be used as an edge-preserving smoothing operator like the popular bilateral filter, but it has better behaviors near edges. The guided filter is also a more generic concept beyond smoothing: It can transfer the structures of the guidance image to the filtering output, enabling new filtering applications like dehazing and guided feathering. Moreover, the guided filter naturally has a fast and nonapproximate linear time algorithm, regardless of the kernel size and the intensity range. Currently, it is one of the fastest edge-preserving filters. Experiments show that the guided filter is both effective and efficient in a great variety of computer vision and computer graphics applications, including edge-aware smoothing, detail enhancement, HDR compression, image matting/feathering, dehazing, joint upsampling, etc.

Supervisor: Niting Wang, Jingwen Wang

 

Time:16:00  October 15, 2020

Address:MingLi Buliding C1102

Chair: Wei Zhang