Let's start big! Think different on AI
The right approach is important for a successful AI Project
The whole thing as a linear (much simplified) process for capturing the AI story thought!
- Basically, it is about the basis of baseline data a training set, a validation set and a test set is generated. For this purpose, data is enriched, normalized, correlated and evaluated with the correct causality in order to be able to apply algorithms to it.
- Depending on the knowledge of the initial data (few or many case studies), the strategy of supervised learning or unsupervised learning is decided fundamentally
- Then someone looks for the appropriate algorithm from the set of the respective provider, which basically pursues the following 5 basic features:
- Simple / Multidimensional Classificationb.
- Clustering (unknown)
- Anomaly Detectiond.
- Regression
- Neural networks
- Now you can use the strategy and the chosen algorithm on the basis of the knowledge of the output data to calculate a prediction / classification / abnormality.
- Now the trained algorithm is ready for optimization – either adapting the algorithm, improving the basic data or adjusting the expectation
- The trained model is statistically evaluated for quality (bias, variance, etc) to achieve best predictive quality. What is the best? Partly not even the man-made quality, but better …. Depending on the source material to data.
That would be the simple run through an AI project. Who noticed it: 80% data preparation / 20% optimization and rest. That’s an approximate ratio of the effort in such projects. Therefore, the relationship between data and their causality or their correlation must be distinguished. Often a connection with data can be found that lies outside of the „first estimate“. A long way to the truth, but more exciting!