It is not easy to be a good thesis tutor. Come and have a look with me.
It is enough to read this article!
I can especially understand the feelings of the questioner, because many students may have the following needs.
1, there are thesis requirements for graduation, but the boss is not very helpful.
Although there is no paper requirement for graduation, I hope to improve my competitiveness in the workplace by publishing papers.
3. I hope to strive for doctoral opportunities abroad by doing research and publishing papers.
After defining the topic, we can continue to decompose the problem into three sub-problems:
1. How to choose a topic? How to learn 3. How to contribute?
1, how to choose a topic
First, choosing the right research direction is half the battle. Don't choose your research direction because of your interest. Because to do independent research, we must first avoid all heavy equipment, such as deep learning. As an individual, you can't beat the group army-you don't have that much computing resources.
Second, choose the direction that suits you. Most basic subjects, such as mathematics and physics, need years of knowledge accumulation and guidance from tutors, because a tutor can save a few days or even months of daydreaming with a touch.
The third point of the topic is to read several classic articles in this field and try the water depth. In other words, it depends on whether you can roughly understand how big the knowledge gap is and how far it is from being able to write independently.
If there are many formulas derived from articles in a certain field and your mathematical ability is limited, then it is not recommended to choose this direction. Reading summary articles is generally a good idea, so that you can quickly see the boundaries of the field and help narrow the scope of the topic.
Of course, interest is still the source of everything, and it is also the backing that can constantly inspire you.
To sum up, topic selection is a balanced process, which is the product of hardware resources+knowledge background+personal interests. If any of them is an absolute short board, it will easily affect the final output.
To be comprehensive, you should choose the direction you are interested in, have some relevant knowledge, have low requirements for resources, and the writing level is not far from the field papers.
2. How to study
When we have a suitable topic, we should first read the classic textbooks or review articles in this field.
My personal experience is: write down your wild ideas while reading, don't think about maturity, write them down. Take a look and see if it can be combined with other fields, such as using ensemble learning or graph mining as a recommendation system.
When reading, narrow down the scope of your own topics and find the topics (chapters) that you are better at through reading.
Most textbooks are divided into chapters, and the chapters are independent to some extent. So you can focus on yourself and read interesting content to learn more.
Suppose we decide a small topic: how to improve the "robustness" of "recommendation system" by using "integrated learning".
So it is not difficult to find a new direction. You need to find benchmark data sets commonly used in this field, and find other baseline algorithm implementations, generally searching for algorithm names on GitHub. You can find relevant works of recent related papers to track the progress in this field.
Find a textbook for comprehensive study. The first step is to reproduce the performance of benchmark algorithm on common data sets, which will be an important reference for research. If some benchmark algorithms are not ready-made, you can try to write one-the process of algorithm realization is often the process of finding inspiration for improvement.
After the above steps are completed, we can consider how to improve the recommendation system with integrated learning. At this time, we can refer to the comprehensive learning textbook, analyze the advantages and disadvantages of different algorithms, find out which methods are helpful to improve robustness, and then apply them to the recommendation system. It is often easier to innovate across fields than in specific fields, especially suitable for independent researchers.
3. How to contribute?
First of all, journals and conferences are generally submissions. Journals are generally comprehensive, but conferences are generally cutting-edge, and different fields care about different things.
Journals are generally blind (that is, the reviewers know your identity and you don't know who the reviewers are). Meetings may be single-blind, double-blind or even triple-blind (such as ICDM). Considering that independent research has no boss endorsement, try to avoid blind submission, because it may suffer.
The second point is to consider the review cycle. The review of most conferences takes 1-3 months, while the first round of opinions of most journals takes more than 3 months to appear. Therefore, if time is sensitive, it is recommended to give priority to meetings rather than periodicals.
Another common operation is that conference papers are submitted to journals after the new content is published (>: 30%), which can give consideration to timeliness and completeness at the same time.
Choosing the contribution channel is also a very unfriendly link for beginners. It is recommended to ask more senior people around you.
The premise of all this is that your English is good enough, which is the premise of all this. The best situation is to cooperate with others, even if it is a novice like you, it doesn't matter. After all, many hands make light work, and there is a psychological support. One more thing, novices who publish independently are more likely to be rejected in the case of single blindness. Adding 1-2 * * * co-authors helps to alleviate this prejudice, because I won't go into details for obvious reasons.
Rejection of papers is very common, and it is inevitable as an independent novice.
4. Summary
Theoretically, as long as you choose the right direction and your own conditions are acceptable, you can always get on the right track after persistent experiments, writing, submission, rejection, revision and re-voting. If you are lucky, you may be able to open up a little situation in your own small research field and have a little popularity.
The greatest sense of accomplishment of independent research comes from "independence". In this process, you will constantly doubt yourself and even deny yourself, which is why I suggest that you can go on with one person.
But when you make progress, such as publishing the first good article, you will be very excited.
Because you have completed the introductory doctoral training, avoided the folk scientific research, opened up the situation in a difficult environment, and even slightly promoted the scientific development. This is more meaningful than publishing the paper itself. You should be proud of yourself.
I am a royal academic leader, and I will continue to share my postgraduate dry goods, scientific research knowledge and thesis writing skills. Welcome everyone to pay more attention! ! !