One of the important process in building AI model is better understanding the project and its scope. How it benefits the society or the company?. Comparing the model with any solution if currently in use.
Not every Solution requires AI😃.
After analyzing scope, Data collection is the next step. data can be collected from different sources based on the model you want to develop. These data are noisy. For getting best results we must clean them before use. Next step is training the model with the data collected. Then the model is tested with the unseen data and to improve the accuracy of the model there are certain views improvement can be done.
Model-Centric AI
Here you collect the data and tune the model to best fit data collected. ie you hold the data fixed and iteratively change the model and re-train the model
Data-Centric AI
Here the data is iteratively improved to give the best results that can be obtained. ie you can hold the model and iteratively gain insight from data and re-train the model.
Which view is best ?
Based on the recent research made by Andrew Ng and deeplearning.ai data-centric view gives better result than the model-centric view. The results are consolidated in the table below
From the table it is evident that data-centric view gives promising improvement than model-centric view. Improved result also depend on the how good the data is preprocessed.
Reference : A Chat with Andrew on MLOps: From Model-centric to Data-centric AI
This is my first post so ignore any mistakes and topics for future blogs are always welcomed 🎉😉
