how does synthetic data function?
In: 0
In the world of AI/ML systems need to be “trained” using data to be able to predict things reliably. Let me illustrate the need for synthetic data through an imaginary example:
Let’s say you are training a system to detect apples in an image. You would train this system by showing it thousands of pictures with apples in them – apples on a tree, apples in baskets, apples on a table etc.
You’ve trained your system using all of these images and deploy your system in the real world. However, you find out that your system is having trouble detecting “apple on a car seat wearing a seat belt”. You realize that very small amount of your “training” images covered this scenario. You try finding more images of apple wearing seatbelt but you can’t turn up much.
In this case you would create “synthetic data” meaning you will hire a graphics designer to “create” several images of apple wearing a seatbelt which look realistic enough. You would retrain your system with your training images + synthetic data and voila! It can now recognize apples wearing seatbelts more accurately!
Latest Answers