This is the second blog.The last one took me nearly two hours, it's costly.I will not talk too much about my daily life, interesting as it is. Due to my restrictions on the use of smartphone, I get free time to read light books, and I'm much more energetic.
In the morning, being late again, I start my business at about 9:20. I'm still devoted to those two papers. I go through paper A, try to copy it into my head. When it comes to Algorithm, I associate the result, the steps with Xiaoyang's comment that the result is not good enough, and understand why the upper bound is in such form. Then I come up with an idea to optimize the result, which is change the way each node stores the sequence to transmitting to affiliated nodes(inspired by that dissemination time is related to the depth of binary tree which arranges the sequence of nodes to attain the file(I refer to those nodes as affiliated nodes), if I can reduce the time of transmitting among affiliated nodes, the total time is shortened). Instead of using binary tree to arrange the sequence, I plan to use Chain originally, then I decide to make short cut, finally power law graph comes naturally. If the diameter of power law graph is smaller than binary tree it's good;if not, nonsense. In the stationary network, it wouldn't help in fact, since both ways need log(n).However, in evolving network, things might be different. If the affiliated nodes are evolving either, the way i proposed needs only constant steps. since there are nodes whose degree are log(n), their load is very heavy. To make it practical, the solution is that for nodes whose degree is larger than certain value, use binary tree to sequence it and ask parent node serve child nodes rather than serve them by itself. Luckily, with high probability the degree is bounded by a constant in evolving power law graph.
In the afternoon, I start to think how to apply the model in B to A. Specifically, how to generate social links by using the model. Graph G in paper B is to be used, but it's generated from graph B which is a bipartite graph having two sets. For present, since I will consider interest in the long run, I assume the two sets in B to be interest set and user set. Xiao yang remind me to try to deduce the two papers, from assumptions to proof to results. it's of great significance. I spent lots of time understanding the algorithm perfectly.
in the night, I improve the method proposed in the morning, and redo the proof of the upper bound.it turns out that i understand the reason. From this experience, I found that drawing an example to explain the process(algorithm), making it vivid will help a lot. Time is not fully made use of, but it's much better compared to yesterday.
Good job and carry on!
2016/8/24
最新推荐文章于 2023-02-21 16:50:08 发布