AI Model Development Phases

[coolharsh55]

Following #82 adding ai:ModelTraining, this issue collects additional concepts for AI development phases such as data collection, data preparation, data labelling, and model fine-tuning.

[coolharsh55]

@DelaramGlp I have the below concepts from generic uses of these terms. I am thinking we can have ai:ModelLifecycle for models which is distinct from ai:LifecycleStage for an AI system.

Data Collection	Collection of data for training or operating AI models
Data Annotation	Annotation of data for training or operating AI models
Data Preparation	Preparing data for training or operating AI models
Model Development	Processing which contributes to development of AI models
Model Training	Processing of data to train an AI model
Model Fine-Tuning	Processing of data to fine-tune an AI model

[coolharsh55]

I think what we have called "lifecycle stages" should instead be called "development phases" to keep it distinct from the ISO and legally defined lifecycle stages. It should also be aligned with the rest of DPV concepts, similar to how we have ai:Training as a type of dpv:Processing.

To support this, we should have ai:ModelDevelopmentPhase as a subclass of dpv:Process, and indicate specific processing activities involved in each stage:

1. ai:DataCollectionPhase --> involves dpv:Collect processing operation
2. ai:DataAnnotationPhase --> involves dpv:Annotate (new, parent: dpv:Generate) processing operation
3. ai:ModelDevelopmentPhase --> involves ai:Technique to produce a ai:Model that will be trained on data in the next phase
4. ai:ModelTrainingPhase --> involves ai:TrainingData and ai:Model as input and produces ai:TrainedModel
5. Testing and validation follow similarly from respective data categories and input/output models.

In addition to the above, ISO/IEC 5338 AI system life cycle processes also has specific "phases" (though mentioned as stages) in between the lifecycle stages that should be added to the above.