| 查看: 3457 | 回复: 13 | ||
| 【悬赏金币】回答本帖问题，作者shangu111将赠送您 500 个金币 | ||
| 当前只显示满足指定条件的回帖，点击这里查看本话题的所有回帖 | ||
[求助]
TKDE 两个major revision 一个reject，最后意见为reject，想申述，请大神看看可能性 已有2人参与
|
||
|
去年投了IEEE Transactions on Knowledge and Data Engineering，一审花了一年，review 1 给的reject， 我觉得绝大部分意见都是错的，有些我专门用了一章节介绍，有些我根本没做过那些假设， 有些是很正常的常识，所以很气愤，感觉完全不尊重别人的成果。我接下来贴出review 1 的意见和我的简单的rebuttal， 想要大家帮我看看申述的可能性。还有想问问有没有人向TKDE申述过。 The issues of this paper are discussed as below: 1. This paper claims that the trading behavior is a function of data cost. But they just take 1(accept), 0(otherwise) into consideration, which do not show any practical correlation with data cost. This contribution is weak and limited. Rebuttal: trading behavior 指的是卖(0)或者不卖(1), 没有其他可能，trading behavior 都可以formulate成data cost 的一个function了，他们之间还没有correlation么？ 2.Compared with the previous works, the authors not only protect the data’ but also protect the trading behaviors’ privacy under LDP framework. However, since the trading behaviors only have two kinds of results 1(accept), 0(otherwise), the real trading behaviors can be obtained by the collectors after trades. Rebuttal: 从来没有reveal real trading behaviours, 一直reveal的,sellers执行的都是perturbed trading decision。 3.The assumption 1 does not always hold since the data may be in a high dimensional space. And PCA cannot ensure any loss in information when projecting the high dimensional data into a lower dimensional space. Rebuttal：PCA 不就是这个用处么？保持90%或以上的variance （这就是保障没有loss in information） 同时减少 data dimension 4.In order to obtain the unbiased gradient for the data collector, the authors show us that the data owners should provide two auxiliary parameters, which is calculated by the biased gradient, with the collector. Firstly, what is physical meaning of these two parameters? Why the data owners provides these two kinds of information to help collector guess their real data cost? Rebuttal：因为你要data owners 要卖他的data啊，要提供auxiliary parameters是你卖data的前提要求啊，何况提供了auxiliary parameters你的data 也被LDP保护 5.Subsection 4.1 states that the data owners perturb the data by using the perturbed gradient. But why is directly perturbing the raw data inapplicable? The authors should attempt to use their designed algorithm to learn real data cost from the perturbed data instead of redesigning a data perturbation method fitting the pricing rule to help learn the real data cost. Rebuttal: 为什么不直接用已经在文章里面单独列了个section说了，一般的learning algorithm over perturbed data都会对这个perturbed data做一些假设，比如符合什么分布，怎么perturbed的，我这篇文章就假设perturbed data是从我这个perturbed scheme 生产出来的不行么？ 6.In Algorithm 2, there is an assumption that the price follows uniform distribution U[0,2]. There is no reference and explanation to support this assumption. Rebuttal:根本没这个假设！！！ In Eq(1), the pricing rule is published with the help of two parameters ϵ,δ. Is it reasonable for the data collector to get the value of ϵ,δ ? Rebuttal：ϵ,δ是data collector设置的，为啥他不知道？ In experiments, Fig.6 is so confusing. There is only one computation time for one trading in practice. Why the authors the CDF of computation time for one trading? Rebuttal：因为整个learning algorithm的gradients 都是randomized，而且计算gradients还用了external library，我做1000 trails，然后画cdf不是很正常么？ |
回复此楼» 猜你喜欢
欢迎211本科同学，过A区国家线，A区非偏远一本，交叉学科课题组
已经有12人回复
-
成果系统访问量大，请15分钟后再尝试。由此给您造成的不便，敬请谅解。
已经有14人回复
-
281电子信息求调剂
已经有8人回复
-
材料277分求调剂
已经有7人回复
-
求调剂院校
已经有12人回复
-
293求调剂
已经有3人回复
-
化学 0703求调剂 总分293 一志愿211
已经有10人回复
-
320材料一志愿西工大专硕英二数二 有两年光伏工作经验
已经有7人回复
-
材料328求调剂
已经有8人回复
-
304分材料专硕求调剂
已经有12人回复
肥耳朵1895
新虫 (著名写手)
- 应助: 0 (幼儿园)
- 金币: 4511.1
- 散金: 400
- 沙发: 4
- 帖子: 2164
- 在线: 382小时
- 虫号: 26941881
- 注册: 2021-08-04
- 专业: 电力系统
9楼2021-10-30 13:38:33
2楼2021-10-30 12:39:09
肥耳朵1895
新虫 (著名写手)
- 应助: 0 (幼儿园)
- 金币: 4511.1
- 散金: 400
- 沙发: 4
- 帖子: 2164
- 在线: 382小时
- 虫号: 26941881
- 注册: 2021-08-04
- 专业: 电力系统
3楼2021-10-30 12:39:13
肥耳朵1895
新虫 (著名写手)
- 应助: 0 (幼儿园)
- 金币: 4511.1
- 散金: 400
- 沙发: 4
- 帖子: 2164
- 在线: 382小时
- 虫号: 26941881
- 注册: 2021-08-04
- 专业: 电力系统
|
你首先要平静情绪，作者要做的不是跟他吵架，而是解释清楚。审稿人不可能看的太仔细，作者需要反思自己的写作，可能并没有能让读者轻松get到贡献。你目前这个反驳，我估计编辑都不一定能满意。首先要花一两句表明文章的贡献，之后再解释细节。 发自小木虫IOS客户端 |
4楼2021-10-30 12:45:58
30