Create datasets that work
in the real world
Properly curated Machine learning datasets can be the difference between a ML system that works in the real world and one that stays in the lab.
Create robust, trustworth AI systems
Complex AI systems need to be trained for unusual but possible situations in order to be robust.
Make sure that your ML systems can handle the most sticky situations by including important edge-cases in your datasets.
Only train your models on clean data
Gargage-in garbage out. Raw datasets can have a large number of bad samples, due to bad lighting, occlusion, blurred vision etc.
Make sure your ML models are trained only on clean and relevant data samples.