After submitting the manuscript on May 26, I was full of doubts and confusion. I didn't know whether my content would be recognized. I was worried and comforted myself not to think about those things, and to do what I should do with peace of mind. After the paper was submitted, many people asked about the specific situation, and I answered them one by one. In fact, many doctoral students were unwilling to tell others about many things they submitted. Not afraid of jokes. But looking back and thinking about it, it might be right for others not to say it. It was so embarrassing to be rejected, and I was also worried that others would be jealous overtly and secretly, and said some yin and yang things, so I simply didn't say anything. Well, it makes sense. I will try not to say anything when others ask me in the future .
I worked as a defense secretary for two days and learned a lot from the review experts, especially the questions they asked. Although some experts did not understand the direction of some of the respondent students, they were still able to ask very professional questions, and their opinions were not groundless, and they gained a lot. The most important gain is the thinking logic of experts on the overall scientific research. I think any scientific research direction or scientific research problem can be roughly summarized as the following: 1. What is the problem to be solved? Because we are engineering, we focus on solving practical problems and do not study purely theoretical problems, so the first thing we need to understand when doing scientific research is, what problems do I want to solve (outlier detection). 2. What contributions have the predecessors made to address this issue? Because the problem we study must not have been created out of thin air, and the predecessors must have proposed many solutions to this problem. From here, a question can be drawn, that is, what are the defects in the contributions or solutions made by the predecessors ? The predecessors must not have solved the problem raised by the first article 100% perfectly, and there must be some defects or deficiencies. Then, for the defects in the work done by the predecessors, we need to optimize (small innovation points) or propose other solutions (large innovation points). 3. What does my proposed method (model) look like? There must be a theoretical basis, derivation, etc. When introducing your own model, you must introduce it clearly. A picture is worth a thousand words, and a suitable picture will greatly improve the understanding of others. One of the most important things in this part of the model is to be logically self-consistent. That is, why the detection effect can be achieved through the model I mentioned, it must be based. Others need to fully understand my thinking through my model. 4. What experiments did I do to verify the effect of my model? Generally speaking, real data sets are used to compare with the SOTA algorithm, and evaluation indicators such as AUC, ACC, DR, FAR, F1, etc. are used to evaluate the pros and cons of the algorithm. In order to improve the credibility of the experimental results, the parameter settings of the algorithm should be introduced in detail. At present, I use MATLAB to reproduce it myself, and use the ELKI toolbox. Let’s see later, many algorithms still need to be reproduced by myself using matlab, so as to ensure the fairness of the experiment.
The understanding of scientific research is basically like this. In the past, too much attention was paid to the model, ignoring the data set and evaluation method. There is too much lack of work in this part, and it is something that we will work hard to do in the next period of time. To become a domain expert, one must be familiar with the entire process. The workload is heavy and the difficulty is not small, but do it a little bit every day, don't stress too much, don't think too much about other messy things, and just be better every day than the day before.