MVDC intro: methodology for Artificial Intelligence Research
Use MVDC methodology when you need to achieve high-risk results and accelerate the research process in Artificial Intelligent projects.
By the end of this article you will be able to…
- See the whole research process and…
- Define the MVDC blocks for your research project.
What is the MVDC methodology?
It is 3-level methodology that describes the full research process in AI project. This methodology is adapted for small iterations and can be applicable with such Agile frameworks like Scrum. Let “Sprint” be the name of such short research iteration.
The MVDC has 3 level as it should stay simple even for short-term research projects (2–4 weeks) and applicable for long-term research projects (1–1.5 years).
MVDC provide you the structured vision of the whole research process. So you can plan your short iterations and measure you research progress effectively.
As the methodology is formulated around the “Sprint”, its first level describes the pattern of such sprint inside the research project. During each iteration we deal with planning, project vision and its program realisation, also we need to collect out experience. The first level consists of 4 main parts that performed sequentially:
- M: Management
- V: Vision
- D: Development
- C: Capitalisation
Depending of the project stage the distribution of efforts between each of the part may vary.
So, at the beginning of the project(1–3 first sprints) the distribution may be: M-20%, V-60%, R-10%, C-10%.
While in active phase it shifts closer to: M-10%, V-20%, R-50%, C-20%.
The second MVDC level describes blocks of each main part.
For “Management” part they are (ERP): Expectation, Risk and Planning. The research project is characterised by biased expectations and high risks. So their management is important process.
“Vision” part consists of (AHS): Analysis, Hypothesis and Solution. The main part of each research project is the vision of the solution. It should be worked out during: solution and market analysis and research hypothesis testing.
The part “Development” contains (DCTD): Data, Code and Test&Deploy. Each of the research hypothesis should be implemented in code that based on data. And these steps deliver the main result.
Finally, “Capitalisation” involves (CAO): Component, Article, Open-source. It is a good practice for projects team to create separate reusable component. And for research team — create articles and open-source as their implementation.
Finally, the third MVDC level describes detailed content of each block mentioned above. It contains best practices for each of the described step that can be used in each AI research project.
The example for the third level is Experiment Journal for “Solution” and “Code” blocks. In AI projects you should be very accurate with keeping a journal of experiments. For such task you can use mlflow.org open-source platform.
In this article you got familiar with…
- 3 levels of MVDC methodology and…
- Few reasons for using MVDC in your practice.
In the coming articles I’ll describe more reasons that formed this framework as is. Also I’ll dive deeper in the second and third levels.
Sources:
- https://medium.com/chingu/a-short-introduction-to-the-scrum-methodology-7a23431b9f17
- https://medium.com/@chrismanna/an-intro-to-the-crisp-dm-methodology-c58cbe0371a3
- https://mlflow.org
Contacts:
tg: @lyoshamipt
ln: Aleksei Goncharov
fb: Aleksei Goncharov
