Developing service-oriented AI models Ep.1 Differences in the AI model development environment
2022/03/04⏱ 5mins EP.1 INTRO
hello. The topic of this episode is 'Developing an AI model for service' .
Recently, more and more people are developing AI models for the purpose of hobbies, research, and study, gathering alone or through the community. That is one sign that interest in AI is increasing day by day, and more and more people are drawing careers in AI companies. So, through the first 'UP Tech', 'how to develop a customer-oriented AI model in the real field' is introduced.
This theme consists of a total of four episodes.
Before getting into the topic in earnest, EP. In 1, I will briefly introduce the difference between 💡 ' AI model development in a controlled situation such as a school/research environment ' and 'AI model development in the real world' . In the following three parts, we will go into detail about the AI model development process and how AI organizations are structured.
AI MODEL DEVELOPMENT IN RESEARCH ENVIRONMENT AND ACTUAL FIELD
📍 What is modeling (model development) in the AI development process?
👉 Utilizing the given training dataset for AI model training, it is to secure the highest performance AI model by verifying the determined test dataset through multiple testing steps. Please refer to [Figure 1] below.
✔️ What is a dataset ? It is a 'collection of data into related units for a specific task', sometimes referred to as a data file or database.
[FIGURE 1] AI MODELING PROCESS
📍 When developing an AI model, what is the difference between the research environment and the actual field?
👉 To start with, it is the difference between a controlled situation and an uncontrolled situation .
AI MODEL DEVELOPMENT IN A RESEARCH ENVIRONMENT IS GIVEN A TRAINING DATASET, A SET TEST DATASET, AND A TEST METHOD IN A CONTROLLED SITUATION. IN OTHER WORDS, IT IS SOMEWHAT FREE FROM UNEXPECTED UNEXPECTED SITUATIONS THAT MAY OCCUR WHEN AI MODELS ARE APPLIED TO ACTUAL SERVICES.
HOWEVER, IN PRACTICE, THERE ARE MANY CASES WHERE NO DATASET IS GIVEN. IN ORDER TO DEVELOP A MODEL, YOU HAVE TO START BY BUILDING A DATASET YOURSELF, AND YOU WILL FACE UNEXPECTED AND UNEXPECTED SITUATIONS IN AN UNCONTROLLED SITUATION. IN ADDITION, THE AI MODEL DEVELOPMENT PROCESS BECOMES MORE COMPLEX AND DIFFICULT BECAUSE THE REQUIREMENTS OF THE CUSTOMER WHO COMMISSIONED THE AI MODEL DEVELOPMENT MUST BE REFLECTED IN DETAIL.
For example, an engineer will receive these 'service requirements' from a customer. “We want to create a service that delivers the XXX experience to our users. Can such a service be implemented with AI?” Then the engineer should start developing the AI model by specifying these abstract requirements to a developable level.
EP.1 GOING OUT
BRIEFLY, WE EXPLORED WHY THE AI MODELING PROCESS DIFFERS IN A RESEARCH ENVIRONMENT AND IN THE REAL WORLD.
✔️ Modeling in a research environment : Creating an AI model with better performance through a given training data set, a set data set, and a test method
✔️ Modeling in the real field : Creating an AI model that meets the customer's service requirements. At this time, the training dataset, test dataset, and test method may or may not be provided depending on the situation.
IN THE NEXT PART, I WILL INTRODUCE IN DETAIL HOW ENGINEERS PREPARE A TRAINING DATASET BASED ON THE ABSTRACT REQUIREMENTS OF A CUSTOMER. IF THERE IS NO GIVEN TRAINING DATA SET, WHAT CONDITIONS AND PROCESS SHOULD THE TRAINING DATA SET BE SECURED TO TRAIN THE AI MODEL? IF YOU WANT TO HEAR THE LIVE VOICE OF THE SCENE, DON'T MISS THE NEXT EPISODE!