If a synthetic dataset repeats the same hour of car noise, it is possible for a learning algorithm to overfit to that hour of car noise and generalize poorly to a new audio clip with different car noise. Even 1,000 unique hours of car noise may cause a similar problem if they come from only 10 cars, because the algorithm may overfit to those cars and perform poorly on audio from a different car.

Overfitting to Reused Synthetic Car Noise

A computer vision system trained on synthetic car images from a video game will probably not work well even if the images look realistic. If 100,000 training examples all come from about 20 car designs, the system will overfit to those car designs and fail to generalize well to dev/test sets that include other car designs.

Overfitting to a Few Synthetic Car Designs

When generating synthetic training examples, the simulated elements must be sufficiently diverse. If artificial data is synthesized using a small set of patterns—such as repeating the same one-hour clip of car noise across all audio samples—the model may overfit to those specific patterns and fail to generalize to new, real-world variations.

Google

Artificial data synthesis is a technique used to address data mismatch by artificially modifying training data to better reflect the development and test sets. For example, if a speech recognition model struggles with background noise, developers can synthesize training examples by adding recorded car noise to clean audio, efficiently creating a large dataset that matches the target distribution.

Learn Before

Related

Learn After