Researchers vs Practitioners

Author(s): Enos Jeba

Originally published on Towards AI.

Computer Vision began with research. We still have research going on but at the same time, some research is mature enough to be implemented into real-world applications.

This also does not mean that research can be stopped. We still have a lot of moon rocks to look under.

This situation where we can also do research and implement the research in the real world provides us with two roads in Computer Vision. Let’s look at how both the roads seem to be.

Research

Idea Focused

Research is motivated by an idea. It mostly goes on with a crazy impossibility and finding all the information required to conclude why it is impossible and then cleverly finding out the way to make it possible.

You will be exploring nodes of math, simulating thoughts or equations, and coming up with neural network architectures for faster or optimized approaches.

Creator

You will be creating neural network architectures.

In research, you can design a neural network that would solve a specific problem faster. E.g. CNN is good with Images.

YOLO is another example of a research paper made for Object Detection. You only look once focused on real-time object detection with the ability to estimate objects in just a frame.

Publish Research Papers

Your idea will be documented as a research paper, which you can publish at a university.

Companies also have research groups that publish research papers, and if they do something unique, they proceed to patent it.

A research paper is commonly funded by Universities, but companies also invest their fair share in research to improve their product.

Research Methodology

The process of bringing ideas to paper is based on a set of rules expressed by Research Methodology.

It includes all the steps with the aspect of how you approached the idea, how you got the data, how was the data/information used, and how all the steps were integrated to arrive at the solution.

It includes all the important aspects of research, including research design, data collection methods, data analysis methods, and the overall framework within which the research is conducted.

Practitioners

Application focused.

Practitioners focus on bringing the networks created by researchers to be put into use in the real world.

If a researcher has created a good face detection algorithm, as a practitioner the focus is on how to make the model available for the desired audience to use it.

Use Neural Networks

Here neural networks are used to be trained on custom data based on subjects of interest in the real world.

Real-time data is utilized to train the model and perform prediction on real-time or static methods.

As it is now a real-world-based entity, there will be an extension for giving alerts or any other action to notify the findings by the model inference to activate a real-life procedure. This is where automation comes into play.

Automation

Let’s say we have trained a neural network to detect faces. If a detected face matches the data in the model, It can mark your attendance or open the door for you to enter. You can also unlock your phone, read the blog below

So now we have these use cases that need to notify or do something when a certain condition occurs on inference, We need to add automation for the next steps.

If a traffic camera captures a person without a helmet on a bike, It should notify the officials.

If a stranger enters unknown premises, it should raise an alarm. If a defect is encountered, the machine should be stopped. These are a few uses that emphasize the automation required.

Implement Research Paper

Being a practitioner, the focus is to implement researched ideas in real life to extend the possibility of the idea to reality.

Data Science Life Cycle

Since the model will be put to use in real life, we have a Data Science Life Cycle to define the stages to implement and deploy the model.

The Data Science Life Cycle starts with understanding Business problems or requirements. The problem may be from different domains like manufacturing, street side, cooperate offices, etc.

Once the problem is understood, we move to the Data Preparation phase. We start by understanding the data available and how to access it.

E.g. CCTV cams provide IP feed or RTSP.

The relevant data will then be collected and validated. Once the data is ready, we move on to data preprocessing. After this, the model will be trained and tested.

The model will then be deployed based on the business requirements.

Which one to choose?

Each road has its ups and downs. One research focused on a problem, while the other focused on the uses of the research produced.

Practitioners get to work with people who are from another field such as business, management, or any other field. The people you will mostly interact with are non-technical people, so you will face explainability problems. To convey your idea, you will have to use layman’s terms, and the technical specification you have probably goes above your head. There are people from other fields who have sound technical knowledge (or some with at least surface-level knowledge) but those people are rare.

Researchers usually have a company of intellectuals around them. They will be equally challenging. You cannot utter an error in the technical terms as you will have a quick consequence of corrections flowing up. Mostly everyone involved would understand your work and you have less need to translate your thoughts into layman’s terms.

If you are lucky enough, You could surely try both together.

Both the roads go hand in hand. If you get into research, you will be also thinking about usability and if you choose the other you will anyway dig your head into research papers and will have exposure to that field also. If you have to implement a unique real-world solution, You will have to put on your thinking cap and become a Researcher. If you want people to find your Research useful, you have to think about its usability.

So they coexist, choose what’s comfortable and you are going to do the other anyway 🙂

Conclusion

The field of Computer Vision offers two distinct yet interconnected paths: research and practice.

Researchers are the trailblazers, pushing the boundaries of what’s possible with innovative ideas, Practitioners, on the other hand, are the implementers, bringing these groundbreaking concepts into the real world

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI