2021年第13期Seminar——黄金诚、徐舒琳

2021年第13期Seminar——黄金诚、徐舒琳

Talk 1

Title: Adaptive Graph Convolutional Neural Networks

Speaker: Jincheng Huang

Abstract:

Graph Convolutional Neural Networks (Graph CNNs) are generalizations of classical CNNs to handle graph data such as molecular data, point could and social networks. Current filters in graph CNNs are built for fixed and shared graph structure. However, for most real data, the graph structures varies in both size and connectivity. The paper proposes a generalized and flexible graph CNN taking data of arbitrary graph structure as input. In that way a task-driven adaptive graph is learned for each graph data while training. To efficiently learn the graph, a distance metric learning is proposed. Extensive experiments on nine graph-structured datasets have demonstrated the superior performance improvement on both convergence speed and predictive accuracy.

Supervisor: Kaiwen Huang

Talk 2

Title:Modeling the Background for Incremental Learning in Semantic Segmentation

Speaker: Shulin Xu

Abstract:

Despite their effectiveness in a wide range of tasks, deep architectures suffer from some important limitations. In particular, they are vulnerable to catastrophic forgetting, i.e. they perform poorly when they are required to update their model as new classes are available but the original training set is not retained. This paper addresses this problem in the context of semantic segmentation. Current strategies fail on this task because they do not consider a peculiar aspect of semantic segmentation: since each training step provides annotation only for a subset of all possible classes, pixels of the background class (i.e. pixels that do not belong to any other classes) exhibit a semantic distribution shift. In this work we revisit classical incremental learning methods, proposing a new distillation-based framework which explicitly accounts for this shift. Furthermore, we introduce a novel strategy to initialize classifier’s parameters, thus preventing biased predictions toward the background class. We demonstrate the effectiveness of our approach with an extensive evaluation on the Pascal-VOC 2012 and ADE20K datasets, significantly outperforming state of the art incremental learning methods. Code can be found at https://github.com/fcdl94/MiB.

Supervisor: Pan Li

 

Time:16:00  January 6, 2022

Address:MingLi Buliding C1102

Chair: Pan Li

2021年第12期Seminar——沈力峰、谭浩

2021年第12期Seminar——沈力峰、谭浩

Talk 1

Title:Suppression of epidemic spreading in time-varying multiplex networks

Speaker: Lifeng Shen

Abstract:

Suppressing and preventing epidemic spreading is of critical importance to the well being of the human society. To uncover phenomena that can guide control and management of epidemics is thus of significant value. An understanding of epidemic spreading dynamics in the real world requires the following two ingredients. Firstly, a multiplex network de- scription is necessary, because information diffusion in the virtual communication layer of the individuals can affect the disease spreading dynamics in the physical contact layer, and vice versa. The interaction between the dynamical processes in the two layers is typically asymmetric. Secondly, both network layers are in general time varying. In spite of the large body of literature on spreading dynamics in complex networks, the effect of the asymmetrical interaction between information diffusion and epidemic spreading in activity-driven, time-varying multiplex networks have not been understood. We address this problem by developing a general theory based on the approach of microscopic Markov chain, which enables us to predict the epidemic threshold and the final infection density in the physical layer, on which the information diffusion process in the virtual layer can have a significant effect. The focus of our study is on uncovering and understanding mechanisms to inhibit physical disease spreading. We find that stronger heterogeneity in the individual activities and a smaller contact capacity in the communication layer can promote the inhibitory effect. A remarkable phenomenon is that an enhanced positive correlation between the activities in the two layers can greatly suppress the spreading dynamics, suggesting a practical and effective approach to controlling epidemics in the real world.

Supervisor: Cheng Yang

Talk 2

Title:Automated Design of Deep Learning Methods for Biomedical Image Segmentation

Speaker: Hao Tan

Abstract:

Biomedical imaging is a driver of scientific discovery and core component of medical care, currently stimulated by the field of deep learning. While semantic segmentation algorithms enable 3D image analysis and quantification in many applications, the design of respective specialized solutions is non-trivial and highly dependent on dataset properties and hardware conditions. We propose nnU-Net, a deep learning framework that condenses the current domain knowledge and autonomously takes the key decisions required to transfer a basic architecture to different datasets and segmentation tasks. Without manual tuning, nnU-Net surpasses most specialized deep learning pipelines in 19 public international competitions and sets a new state of the art in the majority of the 49 tasks. The results demonstrate a vast hidden potential in the systematic adaptation of deep learning methods to different datasets. We make nnU-Net publicly available as an open-source tool that can effectively be used out-of-the-box, rendering state of the art segmentation accessible to non-experts and catalyzing scientific progress as a framework for automated method design.

Supervisor: Li Gou

 

Time:16:00  December 30, 2021

Address:MingLi Buliding C1102

Chair: Li Gou

2021年第11期Seminar——周静、杨建新

2021年第11期Seminar——周静、杨建新

Talk 1

Title:Conditional Convolutions for Instance Segmentation

Speaker: Jin Zhou

Abstract:

We propose a simple yet effective instance segmentation framework, termed CondInst (conditional convolutions for instance segmentation). Top-performing instance segmentation methods such as Mask R-CNN rely on ROI operations (typically ROIPool or ROIAlign) to obtain the final instance masks. In contrast, we propose to solve instance segmentation from a new perspective. Instead of using instancewise ROIs as inputs to a network of fixed weights, we employ dynamic instance-aware networks, conditioned on instances. CondInst enjoys two advantages: 1) Instance segmentation is solved by a fully convolutional network, eliminating the need for ROI cropping and feature alignment. 2) Due to the much improved capacity of dynamically-generated conditional convolutions, the mask head can be very compact (e.g., 3 conv. layers, each having only 8 channels), leading to significantly faster inference. We demonstrate a simpler instance segmentation method that can achieve improved performance in both accuracy and inference speed. On the COCO dataset, we outperform a few recent methods including welltuned Mask R-CNN baselines, without longer training schedules needed. Code is available:https://git.io/AdelaiDet

Supervisor: Na Chen

Talk 2

Title:A Federated Bidirectional Connection Broad Learning Scheme for Secure Data Sharing in Internet of Vehicles

Speaker: Jianxin Yang

Abstract:

Data sharing in Internet of V ehicles (IoV) makes it possible to provide personalized services for users by service providers in Intelligent Transportation Systems (ITS). As IoV is a multi-user mobile scenario, the reliability and efficiency of data sharing need to be further enhanced. Federated learning allows the server to exchange parameters without obtaining private data from clients so that the privacy is protected. Broad learning system is a novel artificial intelligence technology that can improve training efficiency of data set. Thus, we propose a federated bidirectional connection broad learning scheme (FeBBLS) to solve the data sharing issues. Firstly, we adopt the bidirectional connection broad learning system (BiBLS) model to train data set in vehicular nodes. The server aggregates the collected parameters of BiBLS from vehicular nodes through the federated broad learning system (FedBLS) algorithm. Moreover, we propose a clustering FedBLS algorithm to offload the data sharing into clusters for improving the aggregation capability of the model. Some simulation results show our scheme can improve the efficiency and prediction accuracy of data sharing and protect the privacy of data sharing..

Supervisor: Yanan Huang

 

Time:16:00  December 23, 2021

Address:MingLi Buliding C1102

Chair: Pan Li

 

2021年第9期Seminar——苗波、邱花

2021年第9期Seminar——苗波、邱花

Talk 1

Title:Multivariate Time Series Anomaly Detection and Interpretation using Hierarchical Inter-Metric and Temporal Embedding

Speaker: Bo Miao

Abstract:

Anomaly detection is a crucial task for monitoring various status (i.e., metrics) of entities (e.g., manufacturing systems and Internet services), which are often characterized by multivariate time series (MTS). In practice, it’s important to precisely detect the anomalies, as well as to interpret the detected anomalies through localizing a group of most anomalous metrics, to further assist the failure troubleshooting. In this paper, we propose Inter Fusion, an unsupervised method that simultaneously models the inter-metric and temporal dependency for MTS. Its core idea is to model the normal patterns inside MTS data through hierarchical Variational Auto Encoder with two stochastic latent variables, each of which learns low-dimensional inter-metric or temporal embeddings. Furthermore, we propose an MCMC-based method to obtain reasonable embeddings and reconstructions at anomalous parts for MTS anomaly interpretation. Our evaluation experiments are conducted on four real-world datasets from different industrial domains (three existing and one newly published dataset collected through our pilot deployment of Inter Fusion). Inter Fusion achieves an average anomaly detection F1-Score higher than 0.94 and anomaly interpretation performance of 0.87, significantly outperforming recent state-of-the-art MTS anomaly detection methods.

Supervisor: Yongfang Dai

Talk 2

Title:Adversarial Monte Carlo Denoising with Conditioned Auxiliary Feature
Modulation

Speaker: Hua Qiu

Abstract:

Denoising Monte Carlo rendering with a very low sample rate remains a major challenge in the photo-realistic rendering research. Many previous works, including regression-based and learning-based methods, have been explored to achieve better rendering quality with less computational cost. However, most of these methods rely on handcrafted optimization objectives, which lead to artifacts such as blurs and unfaithful details. In this paper, we present an adversarial approach for denoising Monte Carlo rendering. Our key insight is that generative adversarial networks can help denoiser networks to produce more realistic high-frequency details and global illumination by learning the distribution from a set of high-quality Monte Carlo path tracing images. We also adapt a novel feature modulation method to utilize auxiliary features better, including normal, albedo and depth. Compared to previous state-of-the-art methods, our approach produces a better reconstruction of the Monte Carlo integral from a few samples, performs more robustly at different sample rates, and takes only a second for megapixel images.

Supervisor: Rui Huang

 

Time:16:00  December 9, 2021

Address:MingLi Buliding C1102

Chair: Pan Li

 

2021年第6期Seminar——郭志成、李丽

2021年第6期Seminar——郭志成、李丽

Talk 1

Title:Adaptive networks: coevolution of disease and topology

Speaker: Zhicheng Guo

Abstract:

Adaptive networks have been recently introduced in the context of disease propagation on complex networks. They account for the mutual interaction between the network topology and the states of the nodes. Until now, existing models have been analyzed using low complexity analytic formalisms, revealing nevertheless some novel dynamical features. However, current methods have failed to reproduce with accuracy the simultaneous time evolution of the disease and the underlying network topology. In the framework of the adaptive SIS model of Gross et al. [Phys. Rev. Lett. 96, 208701 (2006)], we introduce an improved compartmental formalism able to handle this coevolutionary task successfully. With this approach, we analyze the interplay and outcomes of both dynamical elements, process and structure, on adaptive networks featuring different degree distributions at the initial stage.

Supervisor: Kexian Zheng

Talk 2

Title: Graph Attention Network with Memory Fusion for Aspect-level Sentiment Analysis

Speaker: Li Li

Abstract:

Aspect-level sentiment analysis(ASC) predicts each specific aspect term’s sentiment polarity in a given text or review. Recent studies used attention-based methods that can effectively improve the performance of aspectlevel sentiment analysis. These methods ignored the syntactic relationship between the aspect and its corresponding context words, leading the model to focus on syntactically unrelated words mistakenly. One proposed solution, the graph convolutional network (GCN), cannot completely avoid the problem. While it does incorporate useful information about syntax, it assigns equal weight to all the edges between connected words. It may still incorrectly associate unrelated words to the target aspect through the iterations of graph convolutional propagation. In this study, a graph attention network with memory fusion is proposed to extend GCN’s idea by assigning different weights to edges. Syntactic constraints can be imposed to block the graph convolutional propagation of unrelated words. A convolutional layer and a memory fusion were applied to learn and exploit multiword relations and draw different weights of words to improve performance further. Experimental results on five datasets show that the proposed method yields better performance than existing methods. The code of this paper is availabled at https://github.com/YuanLi95/GATT-For-Aspect.

Supervisor: Guogen Tang

 

Time:16:00  November 4, 2021

Address:MingLi Buliding C1102

Chair: Ze Kang