被拒的论文之感想

这个结果一直是预料之中的事情，但是当结果真的出现在自己眼前时，心里还是有点淡淡的苦涩。那种想着有奇迹出现的心思彻底被浇灭了。当自己冷静下来，认真阅读审稿意见，有点心疼审稿者。因为，自己的写作水平真是 很差劲。审稿者读懂我的论文已不易，何况还要给出那么多指导建议，我很感谢他们。同时，通过他们的建议，我发现自己的工作还可以有更多的层面来深入研究，悲伤之余又有了喜乐。

以后警醒自己，写出的论文，无论创新点是否优秀，都要保证所写内容能够让读者看懂，读起来比较舒服。逻辑上合理，内容上充实，技术上详细，实验结果准确。一个方面是对自己工作的负责，另一个方面是对读者和审稿者的尊重。不能因为自己的一篇论文，而让读者（审稿者）很闹心。哈哈，继续加油。

在着急毕业的同时，也希望自己能够静下心来，认真的雕琢自己的工作，使得每一个工作都能够担当得起自己付出的时间和心血。谨以此勉励自己，警醒自己。与志同道合的人共勉，愿我们都能成为大写的人。

下面附件审稿意见（看到这些，你该作何想？）：

======= Review 1 =======

> *** Recommendation: Please provide youroverall assessment of the paper.
Likely Accept (This paper should be accepted butI will not champion it) (4)

> *** Expertise: Please indicate your levelof expertise with the topic of the paper.
Basic knowledge (2)

> *** Summary: What are the major issuesaddressed in the paper? Do you consider them important? Comment on the novelty,creativity, impact, presentation, and technical depth.

This paper studies the problem of activityrecognition using CSI information from a single WiFi AP. This is a wellinvestigated, yet still open, research problem. A holistic approach ispresented including coarse-grained activity recognition, detection of start andend of activity, and distinguishing between independent and continuousactivities. These are all important aspects of the general recognition problem. The implemented system and experimentation validate the accuracy ofthe proposed approach and can significantly impact future works in this area.

> *** Strength: Please discuss the mainstrengths of this paper.

1. The paper is well written (with few typosonly outlined below) and structured. The problem is nicely motivated while thedescription of related work is sufficient.
2. Solution methods for different problems arepresented such as the DWT, window-based moving variance and SRC methods.
3. Various practical challenges faced whendesigning a WiFi-based recognition system are taken into consideration. Theseinclude the subcarrier sensitivity, speed, location, height of volunteer, etc.
4. The implementation and experimentationresults are extensive. They clearly show the impact of different systemparameters on the recognition accuracy.

> *** Weakness: Please discuss the mainweaknesses of this paper.

1. A weakness of the paper is that itstheoretical contribution is rather limited. To a large extent, the paper isbased on previous methods for dealing with theoretical activity recognitionproblems, e.g., classic  classification algorithms.
2. The presented results are limited to one APand certain types of rooms. It is unclear if the reported high accuracy ofrecognition would  remain in more realistic conditions.
3. Additional evaluation results that comparewith state-of-the art methods could have been provided to strengthen the paper.
4. Additional explanation is needed about thedifferences between the values in table II.
5. Some language errors in the text are thefollowing:
-previous works faces->previous works face(page 2)
-we proposed CAR->we propose CAR (page 2)
-Therefore, We->Therefore, we (page 4)
-is more robustness->is more robust (page 4)

> *** Detailed Comments: Please providedetailed comments that will help the TPC assess the paper and help providefeedback to the authors. High quality review comments are extremely importantfor the discussion in the TPC meeting.

Please see my detailed comments above.

======= Review 2 =======

> *** Recommendation: Please provide youroverall assessment of the paper.
Likely Accept (This paper should be accepted butI will not champion it) (4)

> *** Expertise: Please indicate your levelof expertise with the topic of the paper.
Familiar (3)

> *** Summary: What are the major issuesaddressed in the paper? Do you consider them important? Comment on the novelty,creativity, impact, presentation, and technical depth.

The paper presents CAR, a CSI-basedCoarse-grained Activity Recognition using a single AP.
It used CSI to detect 16 activities in indoorenvironments by exploiting advanced classification based on DWT. It divides theCSI sequence into multiple segments and use the combination of multiplesegments to increase the classification accuracy. It was provided effective.

> *** Strength: Please discuss the mainstrengths of this paper.

+ It is challenging and also useful to enablecoarse-grained activity recognition using a single AP.
+ The proposed algorithm is effective andoutperforms the state of the art.
+ The work is well written and executed.

> *** Weakness: Please discuss the mainweaknesses of this paper.



> *** Detailed Comments: Please providedetailed comments that will help the TPC assess the paper and help providefeedback to the authors. High quality review comments are extremely importantfor the discussion in the TPC meeting.

I enjoy reading this paper. This is well writtenand executed.
The proposed solutions leverages the combinationof multiple segments to effectively improves the classification accuracy.

Here are several questions regarding itssensitive to data quality.
(1) Will CAR handle samples obtained fromdifferent locations well? CSI should vary at locations. So have you consideredit? Have you tested with training samples from one location and test samplesfrom another location? Without addressing it, CAR’s application is quitelimited.

(2) Have you notice the impact of differentuser? Compared with fine-grained activity, coarse-grained activity can lowerits sensitive to test users with different height and figures. However, haveyou considered them? Is is user-dependent classification or a generic one wheresamples from different users are mixed?

(3) I assume that the paper will consider theuser perform activity at one specific location. No more mobility will beconsidered. Please clarify it. I guess that CAR may work poor under mobility.

Without considering the above factors, CAR maybe still useful but its practicality is limited. It should be addressed in theeval or in the discussion.

======= Review 3 =======

> *** Recommendation: Please provide youroverall assessment of the paper.
Definite Reject (I will fight against acceptingthis paper at the TPC meeting) (1)

> *** Expertise: Please indicate your levelof expertise with the topic of the paper.
Familiar (3)

> *** Summary: What are the major issuesaddressed in the paper? Do you consider them important? Comment on the novelty,creativity, impact, presentation, and technical depth.

In this paper, the authors develop a system calledCAR to recognize coarse-grained activities. In fact, many CSI based activityrecognition schemes have been proposed in recent years. The main contributionin this paper is to segment the continuous activities, which is a practicalissue. However, the lack of several important details makes it confused tounderstand the ideas. Presentation and technical depth are required to improvein this paper.

> *** Strength: Please discuss the mainstrengths of this paper.

1. The authors implement their system underdifferent scenarios and compare the results with different schemes.

> *** Weakness: Please discuss the mainweaknesses of this paper.

1. In this paper, one main contribution is tosegment activities. To detect the direction of an activity, the authors claimthat they explore the difference between antennae on the signal pattern.However, they do not explain how to parse direction information. Meanwhile,related experiment is missing.

2. There are several serious errors inexperiment results. For example, in page 7, the authors claim the accuracy ofNB algorithm is 64%（50%）, which is not in agreement with the results shown inFigure 11(b)；In table III, the sum of probability of error and correct is not equal to1 for label 3,8 and 16.
3. The presentation of experiment results isvague. For example, in table III, they do not give the correspondence betweendifferent activities and labels. Meanwhile, the authors claim that theactivity13 is recognized with the accuracy error 12.79% as the activity7 withoutanalyzing reasons.
4. The tables and figures in this paper are notprofessional. For example, figure 10 seems a little informal; the bold font intable II has not been explained; the line between label 7 and 8 seems lessmeaningful.

> *** Detailed Comments: Please providedetailed comments that will help the TPC assess the paper and help providefeedback to the authors. High quality review comments are extremely importantfor the discussion in the TPC meeting.

1. In page 5, the authors claim that they explorethe difference between antennae on the signal pattern to detect the directionof an activity. However, they do not explain how to parse directioninformation. Meanwhile, related experiment is missing.
2. In page 6, figure 10 seems a little informal.Meanwhile, the experiment results under different scenarios, with and withoutempty baseline environment are missing.

3. In page 7, the authors claim the accuracy ofNB algorithm is 64%（50%）, which is not in agreement with the results shown inFigure 11(b).

4. In page 7, the bold font in table II has notbeen explained.

5. In page 8, the sum of probability of errorand correct is not equal to 1 for label 3,8 and 16 in table III.

======= Review 4 =======

> *** Recommendation: Please provide youroverall assessment of the paper.
Definite Reject (I will fight against acceptingthis paper at the TPC meeting) (1)

> *** Expertise: Please indicate your levelof expertise with the topic of the paper.
Basic knowledge (2)

> *** Summary: What are the major issuesaddressed in the paper? Do you consider them important? Comment on the novelty,creativity, impact, presentation, and technical depth.

The paper introduce a Channel State Information(CSI)-based Coarse-grained Activity Recognition system named CAR. Authors havethe objective of finding the signal pattern corresponding to the specificactivity.  The evaluation is conducted through experiments. Experiment
results with one user show that CAR systemachieves an average accuracy
of 93.35%

> *** Strength: Please discuss the mainstrengths of this paper.

Use of the joint method of Discrete Wavelet(DWT) and window-based moving variance to determine the precise start and theend of activities data corresponding to the activity.

Design a method to distinguish an independentactivity from a continuous activity and classifier based on the sparserepresentation classifier (yet already used in recent works).

Implementation of the system in differentsettings.

> *** Weakness: Please discuss the mainweaknesses of this paper.

The ideas are presented chaotically and are hardto follow.

Using variations of single sub-carriers todescribe the human activity is not new.

The paper is overall poorly written.

> *** Detailed Comments: Please providedetailed comments that will help the TPC assess the paper and help providefeedback to the authors. High quality review comments are extremely importantfor the discussion in the TPC meeting.

In Figure 2, it is unclear what carrier of theCSI is plotted; or are the results the output of some filter over multiplecarriers?

Results in Fig 3(b) are expected by the spectralcoherence of the channel.

There are a few typos, such as "An activityis consists of...", "We collect data is real-time","according the J value", "The following works, we willexplore",...

Section II.3 about the design framework isconfusing and a repetition of things that have been already said in previoussections.

======= Review 5 =======

> *** Recommendation: Please provide youroverall assessment of the paper.
Likely Reject (This paper should be rejected butI will not fight strongly against it) (2)

> *** Expertise: Please indicate your levelof expertise with the topic of the paper.
Expert (4)

> *** Summary: What are the major issuesaddressed in the paper? Do you consider them important? Comment on the novelty,creativity, impact, presentation, and technical depth.

This paper proposes a CSI-based coarse-grainedactivity recognition system in indoor environments.
The proposed scheme leverages a series of novelmethods to synchronize, segment and classify different activities and can beused to monitor continuous activity.
The authors claim that their work can achieve anaverage accuracy of 93.35% in the coarse-grained activity recognition.

> *** Strength: Please discuss the mainstrengths of this paper.

(1)The idea to monitor continuous activityrather than identical activity to make the state-of-the-art advances moreapplicable in practical use is impressive and convincing.
(2)The proposed methods seem promising andprogressive.
For example, the authors propose a joint methodof the DWT and window-based moving variance to determine the start and end ofan activity.
(3)The claimed evaluation result looks good andseems to be a step further to the existing works related to CSI-based activityrecognition.

> *** Weakness: Please discuss the mainweaknesses of this paper.

(1)The authors should focus more on thecontributions of their work.
For example, to detect the boundaries ofactivity, the joint method of the DWT and window-based moving variance ispresented as a fairly straightforward endeavor.
It would vastly help the reader to introduce thedesign decisions and choices made.
(2)The author claims to have designed a methodto detect continuous activity, but have not introduced detailedly.
(3)In the segmenting part, the author mentionedthat various directions of sub-activities produce a slight change in the signalpattern, but didn't introduce how the pattern is like,
which makes this solution less convincing.
(4)From the above 1~3, the contribution isover-claimed, this work seems to be a combination of existing methods andintuition.
(5)The recognition process seems to be doneoff-line, which makes the system less applicable.

> *** Detailed Comments: Please providedetailed comments that will help the TPC assess the paper and help providefeedback to the authors. High quality review comments are extremely importantfor the discussion in the TPC meeting.

The quality of writing is poor. There are toomany vocabulary errors and grammatical errors.
For example, "An activity is consists ofsome ..."in page 2, "we collect data is real-time andaccidental..." in page 3.
The errors are too numerous to count, and evenreduce readers' comprehension.
The paper put too much weight on backgroundinformation rather than the technical details, making the proposed solutionsless convincing
The observations introduced in part II.A seemsto be presented just to increase the contributions of this paper.
I can not find any relationship between theseobservations and the solutions proposed.
The segmenting activity algorithm is vague andshould be introduced more detailedly.