1.赵子豪：《Experimental Analysis of Distributed Graph Systems》
2.Fayaz Ali： 《A Review of Researches on Deep Learning in Remote Sensing Application》
3.乔子越：《Gated Graph Sequence Neural Networks》 ICLR’16
1.《Experimental Analysis of Distributed Graph Systems》
Graph processing systems:
- GraphLab (PowerGraph)
- Flink Gelly
- GraphX (SPARK)
- World Road Network
- UK 200705
- Vertex-centric:Synchronous: Giraph , GraphLab , BlogelVertex (Blogel-V) 
2.1 Vertex-Centric BSP
Each vertex computes its new state based on its current state and the messages it receives from its neighbors. Each vertex then sends its new state to its neighbors using message passing. Synchronous versions follow the Bulk Synchronous Parallel (BSP) model that performs parallel computations in iterative steps, and synchronizes among machines at the end of each step. This means that messages sent in one iteration are not accessible by recipients in the same iteration; The computation stops when all vertices converge to a fixpoint or after a predefined number of iterations.
Giraph is implemented as a map-only application on Hadoop. It requires all data to be loaded in memory before starting the execution. Graph data is partitioned randomly using edge-cut approach, and each vertex is assigned to a partition.
GraphLab is a distributed graph processing system that is written in C++ and uses MPI for communication. Similar to Giraph, it keeps the graph in memory.
It has three functions: Gather, Apply, and Scatter (GAS). The GAS model allows each vertex to gather data from its neighbors, apply the compute function on itself, and then scatter relevant information to some neighbors if necessary.
This replicates vertices and helps better distribute the work of vertices with very large degrees.
Blogel adopts both vertex-centric and block-centric models.
Blogel is implemented in C++ and uses MPI for communication between nodes.
2.2 Vertex-Centric Asynchronous
2.3 Block-Centric BSP
Blogel-B: 对一个block有一个serial graph algorithm。使用GVD（Graph Voronoi Diagram）把数据集分成多个连通子图。
2.5 MapReduce Optimized
- 在vertex table上更新多个值开销比创建新表更大。超过阈值就创建新表，而非修改值。
2.7 Stream Systems
定义操作，形成拓扑图，数据在拓扑图中被处理、传递并形成数据流。 本文分析了Flink Gelly。两种approach：stream和batch。为了保证和其他工具一致，文中使用的是batch，这样可以把准备数据的时间和计算时间分开。
weakly connected component
Single Source Shortest Path
4. Experiment Design
|Systems||Giraph, Blogel, Hadoop, HaLoop, GraphX, GraphLab, Vertica, Flink Gelly|
|Workloads||WCC, PageRank, SSSP, K-hop|
|Datasets||Twitter, UK, ClueWeb, WRN|
|Cluster Size||16, 32, 64, 128|
Amazon EC2 AWS r3.xlarge machine.
4核s, 30.5G内存, Xeon E5-2679 v2, SSD硬盘。
4.2 Evaluated Metrics
|1.46B||35 / 2.9M||5.29|
|WRN||717M||1.05 / 9||48K|
|UK200705||3.7B||35.3 / 975k||22.78|
|ClueWeb||42.5B||43.5 / 75M||15.7|
5.1 Blogel：The Overall Winner
因为没有昂贵的“infrasturcture”（such as Hadoop or Spark）。使用高效的C++库，优化利用所以CPU核，有小的内存footprint。
5.2 Exact vs. Approximate PageRank
5.3 Syncchronous vs. Asychronous
5.6 GraphX is not efficient when large number of iterations are required.
GRAPHX/SPARK比所有其他系统都慢。因为spark overhead，data shuffle,长的RDD血统，和checkpoint。以前的研究表明，GRAPHX高效，因为它使用了一个特殊的Spark原型版本，in-Memory的shuffle。此功能在最新版本不可用 GraphX在所有规模的集群上基于WRN做WCC都失败了，因为超过了内存大小，有的是超时。结果证明，spark保持RDD血统的容错机制导致了内存错误。当迭代次数增加时，这些lineage就变成了内存占用的大户，导致了潜在的内存不足错误。
2. 《A Review of Researches on Deep Learning in Remote Sensing Application》
1. Understanding Images:
As Images are the lagest source of data so we need to understand in order to solve major problems.such as sheer size of data/ The problems of search, annotation, and classification of videos on Youtube are examples of challenges where the size of image data becomes a problem.
Directed vs. Undirected Graphical Models There are two types of graphical models: Directed Graphical Model (or Directed Acyclic Graphs- DAG) and Undirected Graphical Model (UGM). The directed edges in a DAG give causality relationships, DAGs are also called Bayesian Network. The undirected edges in UGM give correlations between variables, UGMs are also called Markov Random Fields (MRF). There are two types of ways to organize and represent relationships between variables. Here is an example of representing the regulatory relationships of proteins and genes in both ways.
2. Remote sensing:
It is a technical means using sensors on satellite, aircraft or other platforms to collect targets’ radiation information, with which specific information can be obtained. In recent years, with the rapid development of remote sensing technology, the capacity of acquiring remote sensing data has been enhancing. Meantime, the spectral, spatial and temporal resolution of remote sensing imagery have been improving.
Deep learning is an important domain of machine learning research. Compared with traditional machine learning, deep learning is a representationlearning method with multiple layers. Data abstraction and extraction from the lower layers to higher layers are accomplished through simple nonlinear modules. Current deep learning often use deep neural network (DNN) to construct the layers, which are the stacks of simple nonlinear modules. Input data is passed between the layers, whose mapping relationship reduces the dimension and extract the key characteristics of data . Relying on the deep convolution neural network (DCNN), deep learning provides an end-to-end machine learning model that can automatically extract image features without extraction algorithms designed by human. Compared with traditional methods, deep learning is completely data-driven, which can automatically find the best ways to extract image features through learning.
2.1 Common Deep Learning Methods in Remote Sensing Application:
the deep learning method in remote sensing application is mainly used in three aspects, namely surface classification, object detection and change detection. A review of the current research results indicates that the major technical approachM is to translate specific problems into classification or object detection tasks, which are processed with the computer vision deep learning model that is redesigned and adjusted for the targets of the remote sensing application, thus the specific problems are solved.
Land cover classification is a major field of remote sensing application. The main task of surface classification is to divide the pixels or regions in remote sensing imagery into several categories according to application requirements [4,7]. The deep learning model of land cover classification is generally based on deep belief network (DBN), convolution neural network (CNN) and spare auto encoder (SAE), among which the deep convolution neural network is the most popular approach at present.
Many early studies used deep CNN as Alexnet and VGG Net and achieved certain results. However, the nature of Alexnet and VGG Net classification method is to transform an image into a corresponding eigenvector through convolution, pooling and fully connected layer. Based on the eigenvector, a value representing the image classification is output. Therefore, the major issue addressed with such approach is the classification of integrated imagery on the image level. However, land cover classification is a problem of image segmentation, what to be addressed is the multi-classification after semantic segmentation of a single image
Object detection is another common application of remote sensing. The deep learning model of object detection is mainly based on region-based convolution neural networks (R-CNN), which is the earliest proposed method of deep learning object detection. The main idea is to transform the object detection problem into the classification problem. The image is divided into a large number of candidate regions by selective search algorithm, CNN is then applied to obtain the eigenvectors of candidate regions, and finally object detection is completed by the classifier, which determines the type of the candidate area . The proposal of R-CNN has greatly improved the success rate of image object detection, but R-CNN will generate partially overlapping candidate areas from each detection target. Such areas are repeatedly fed into CNN for feature calculation, thus reducing the efficiency of detection. To reduce overlapping candidate areas, He Kaiming proposed Spatial Pyramid Pooling Networks (SPP-Net) , which introduces the spatial pyramid pooling layer after the last convolution layer, thus repetitive processing is eliminated, allowing image of any sizes to be processed with CNN. With these improvement, SPP-Net has greatly increased the speed of object detection. Based on SPP-Net, Girshick proposed Fast R-CNN , which simplifies the spatial pyramid pooling layer of SPP-Net, thus, the RoI pooling layer is formed to extract features. The substitution of SVM by Softmax greatly improves the speed of training and detection. It is more accurate and 213 times faster than R-CNN. To further improve the efficiency of Fast R-CNN in generating candidate area, Ren et al. proposed Faster R-CNN , which introduces Region Proposal Network (RPN), meantime, RPN and Fast R-CNN are combined as an integrated network to generate candidate regions. With further improved network structure, YOLO  and Single Shot Multibox Detector (SSD)  maintain almost the same detection accuracy with significantly improved detection speed
change detection Change detection is the process of detecting changes using remote sensing imagery obtained at different times. These changes are due in part to natural phenomena, such as droughts, floods, and landslides, the other part is due in human activities as new roads, excavation of the surface or construction of new houses. Compared to models for surface classification and object detection, there are less deep learning models for image change detection . The current change detection based on deep learning mainly adopts two technic approaches. One is to detect the correspondent points of two imagery through deep learning and determine whether there are changes to the correspondent points. The other approach is to translate the change detection problem into the surface classification problem, and acquire the changed region through semantic segmentation, comparing and classification of map spots. From the experimental results, the semantic segmentation approach is easier to achieve, faster in speed and better in detection accuracy.
Gated Graph Sequence Neural Networks
 Li, D.R., Tong, Q.X., Li, R.X., Gong, J.Y. and Zhang, L.P. (2012) Current Issues in
High-Resolution Earth Observation Technology. Science China Earth Sciences, 55,
 Pagot, E. and Pesaresi, M. (2008) Systematic Study of the Urban Postconflict Change Classification Performance Using Spectral and Structural Features in a Support Vector Machine. IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing, 1, 120-128. https://doi.org/10.1109/JSTARS.2008.2001154
 Li, D.R., Zhang, L.P. and Xia, G.S. (2014) Automatic Analysis and Mining of Remote Sensing Big Data. Acta Geodaetica et Cartographica Sinica, 43, 1211-121
 Lecun, Y., Bengio, Y. and Hinton, G. (2015) Deep Learning. Nature, 521, 436-444. https://doi.org/10.1038/nature14539
 Gong, J.Y. and Ji, S.P. (2018) Photogrammetry and Deep Learning. Acta Geodaetica et Cartographica Sinica, 47, 693-704.
 Gong, J.Y. and Ji, S.P. (2017) From Photogrammetry to Computer Vision. Geomatics and Information Science of Wuhan University, 42, 1518-1522.
 Ball, J.E., Anderson, D.T. and Chan, C.S. (2017) A Comprehensive Survey of Deep Learning in Remote Sensing: Theories, Tools, and Challenges for the Community. Journal of Applied Remote Sensing, 11. https://doi.org/10.1117/1.JRS.11.042609
 Badrinarayanan, V., Kendall, A. and Cipolla, R. (2017) SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation. IEEE Transactionson Pattern Analysis and Machine Intelligence, 39, 2481-2495.