Research Theory and Practical Experience

This English version was machine-translated from the Chinese original. Technical terms may require verification.

The methodological basis for research

The questions in this article can also be addressedScientific writing guide、Methodological guide for the synthesis of literatureHow the concept of a relatively close read together is developed in different contexts.

Extensive literature in relevant fieldsIt's very important. We need to comb it.What are the problems in the field?This is how you choose your own research direction.

What we need to learn is:

• The core field is over-heard - it depends on the MIT, B, Web-Opening, MOOC and other platforms.
• For literature, keep it wide (even rough) and read classics and see the work of the best scholars. Classes

We need to...Found papersI'm sorry. In fact, authors tend to show in their articles the better part of what they do, and avoid being bad at doing things. This is the function of recurrence: either testing new data or looking at the same methods used to deal with different issues.

How to solve it? It also depends on reading more and seeing how current scholars are solving problems.

For specific scientific processes, we should test our own ideas on simple data and continue to optimize them based on the results of the tests.It's not scary to meet a workout, trying to analyze the exact reason.And then fix it. The search for specific reasons allows for a comparison of the differences between good and bad data. The algorithm design of the literature helps us see if it's a problem. And finally, don't forget to see if the question we're looking for is right. Finally, it migrates to real data.

The main directions for conducting long-term research are:

• Promoting industrial progress
• Advance algorithms and address the already defined issues raised
• Asking new questions and trying to solve them.

Ideas always emerge when they focus on the links between knowledge and broaden their intellectual landscape. Every little thing that is done is a little bit of a bigger thing, and it's the small thing that is done first, then a big system, and it's the usual thing in the field of scientific research.

A guide to work.

Identity Change

Objective reality:: Increased percentage of autonomous learning, moving from learning old knowledge to researching new knowledge.

What do you want?：

• I'm talking to other people. I'm learning.
• When you touch the strange theory, it's normal to be lost and slow, and it's okay to go around the corner.

Scientific rhythm

First, there is a need to identify individual research interests. After a clear interest, needCurrent situation in the field of researchSee if you can find the right direction. The combination of interest and support for hardware conditions determines whether the topics selected above are to be pursued.

The error is not a terrible one, it is a cumulative and productive one, and it is very common to try.

The rhythm is very important - after losing standardized curricula, balancing reading and experimenting in literature, balancing scientific work and life is very important, and ensuring WLB is efficient.

Paperwriting

It's hard to start, and it's more important to finish than to be perfect.

Basic structure - a clear and clear presentation:

1. Introduction: What are we dealing with?
2. Introduction: How far has this been studied at home and abroad?
3. TextHow do we solve this?
4. Outcome and discussion segment: How far have we solved this problem?

The use of some documentation management tools is excellent and can lead to significant efficiency gains.

Academic reports

The most basic academic report is a grouping exercise, which is also an important means of obtaining feedback from the audience. The design of an academic report and a grouping is completely different.

Academic reports need to consider:

• Audience background
• Interactive Mode
• Factors such as completeness and duration of reporting

The PPT, which is also very important in academic reports -- is the most important thing to be clear and concise.LESS IS MORE。

Writing skills

Workload

• The overall workload must be adequate. This is the trend of journal postings.
• One of the things that is sufficient is that the comparison experiment is sufficient -- to select the 10-20 algorithms of recent years to compare as many indicators as possible, and to ensure that a huge table is available.
• The digestion experiment is also sufficient - more modules are selected to digest and mechanisms that may be removed. If the number of experiments is insufficient, consideration could be given to removing multiple modules from the study portfolio at the same time; if not enough, it would be a sort of a accommodation to consider streamlining the original complex modules. It's three to four for the digestion experiment, of course, as much as possible.

About Data Sets

Scientific data can be collected either from open databases or by researchers themselves. Data sources should not be selected only for size or heat, but should first identify the units of observation, research design, time horizon, sampling mechanisms and available variables needed to study the problem, and then check access conditions, data dictionary and replicability requirements. Open databases are easy to reproduce and compare, and self-build data are easier to fit into particular issues; both must face representativeness, measurement error and selection bias.

Clinical and public health database reference

Database	Research design and data particle size	Access	Main limitations
NHANES	Repeated cross-section surveys; provision of individual-level questionnaires, medical examinations and laboratory data	Open Download	Survey cycles cannot be considered directly as vertical follow-up visits by the same population, and combined cycles require adjustments to the official rules Heavy
UK Biobank	Forward-looking queues; provision of individual-level long-term follow-up and multi-modular health data	Applications for studies	Participants have a selection bias and caution is required when extrapolating conclusions to the general population
CHARLS	China 45 Longitudinal survey of older persons aged 45 and over; including individual and household data	Application for registration	Cross-stage research needs to address tracking missing visits, variable calibration changes and investigative powers Heavy
GBD	Summary data by region, period, disease and risk-based organization model estimates	Open tools and downloads	No individual cases are included, no individual level inferences can be made on this basis and ecological errors must be observed. Error

NHANES uses complex multistage, multistage probability sampling, not simple random sampling. The overall extrapolation is made using both the corresponding sample weights, the layered variables and the main sample unit (PSU); when using a specific inspection or laboratory subsamples, the weights that match the module should also be selected.

In-depth learning modular innovation

Deep learning is a multi-module structure, and there are more serious hydrological phenomena - – Because their work is a stack of modules, and simply stacking different structures is the creation of innovation points, and finally drawing a flair-breathed network structure.

** On a special type of technical: self-adaptation dynamic XX module. ** To cite an example: for the original standard issue, we often use parallel modules in many places (e.g., the element of character extraction, the integration of characteristics) that are the same. Whereas our previous approach to integration was generally equal weight-added, the dynamic module of adaptation is to add a dynamic weighting branch to integration, and to adapt to the weight-sharing and then to name.

No serials are presented here - although the use is high, it usually allows interaction between modules if it is more complex, but this level of modification is more complex and is not presented here.

Writing thinking

1. Change Variable: The original study x1, x2, x3 et al. for y, we can replace the parts x and y to get a new article. This is generally quantitative research.
2. Innovative approaches• A more innovative approach to revisiting the problems that already exist, to arrive at better conclusions. Or build on existing research and innovation models.
3. Switching scenarios: New findings may be available for research based on data available in parts or parts of the world, in exchange for our data sources and research scenarios. A single scenario can be studied or a comparison of different scenarios can be made.
4. Reverse proof• To deny the conclusions of existing mainstream research trends or the general perception of society.

These are only relevant to empirical analysis.

Consolidated ranking and data beautification

1. Reduced sample size, reduced accuracy of estimation, and avoided hypothetical gaps
2. Multi-indicator integrated assessment algorithm to avoid the falsification of a single indicator that requires direct glorification of the indicator
3. Because of the different characteristics of the indicators themselves, we give the individual methods separate rankings based on a single indicator and then develop a final evaluation system based on a secondary ranking
4. Consider using the average, median, first/second largest (number) base mathematical statistical indicators of ranking
5. Consider a combination of credits based on ranking - individual rank ' s credits are determined according to the circumstances, ensuring that the required algorithms are high min
6. Continue secondary development of indicators based on rank, using statistically unsupervised learning extraction indicators
7. These are single data sets, maybe one more time on multi-data integration. - After all, the multi-data set is by nature a multi-perspective indicator.