When one embarks on a journey to develop artificial intelligence capabilities within a system, it quite often begins with the identification and collection of data from relevant sources.
Once data is gathered, next one may need to massage the data to convert it into a form that is ingestible by AI models. For instance, some thing as simple as an address may need to the split into its constituent parts of street number, street name, city, state, zip and country. Such cleansing activities occur at this stage.
The area of Machine Learning is a set of tools that one can employ to derive insights from data. Now, just like a hammer is not a solution to every woodworking project, one has to select which models best fit the problem at hand in Machine Learning.
In Machine Learning, one trains a system to look for patterns. That training process, often involves one exposing the models selected from the previous step to typical training data that the system would most likely encounter is everyday use. The better the training data, the better the outcomes that can be predicted through Machine Learning when it comes to real world scenarios.
Evaluation or Testing involves running various typical data sets against the machine learning models to observe the outcomes predicted. If the outcomes are not accurate, then one may need to feed better training data or tweak the models. This is an iterative process to ensure that in real-life scenarios the predictions made by the AI system are accurate.
Hyperparameters can be thought of as settings that govern the training process itself. So, for instance, in a deep neural network (one of your Machine Learning tools) is a hyperparameter that indicates the number of hidden layers of nodes to use. The process by which these parameters can be adjusted over time by measuring the accuracy of predictions is called hyperparameter tuning.
The ultimate objective of Machine Learning systems is to make predictions. These predictions should not be thought of as end-point but rather as a step in a continuous process. By this we mean that the predictions serve as as a feedback loop to the other layers to help improve future predictions.