Are Synthetic Training Examples Representative?
When creating synthetic training data, check whether the generated examples really resemble the real examples the model must handle. If the synthetic set has obvious shortcuts or narrow coverage, a model may learn to spot the generation artifacts instead of the task itself. For instance, synthetic warehouse photos made from only a few camera angles, or synthetic machine-noise clips built from just one short recording, may be easy to distinguish from authentic data and therefore less useful for training.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
Data Mismatch May Not Have a Clear Fix
Synthetic Data That Approximates the Dev Distribution
Are Synthetic Training Examples Representative?
What should you do when training results are strong but validation results drop because the data sources differ?
A data mismatch problem happens when a model does well on training data but performs poorly on a dev set that comes from a different distribution.
Matching the Dev Set Environment
Match each data-mismatch idea to its description.
Order the steps for diagnosing and fixing a data mismatch in a customer-feedback classifier.
What most likely explains the model’s weak performance on the development set in this speech project?
Does training on examples that look more like the dev set always eliminate data mismatch?
When a speech recognizer performs poorly on noisy clips in the dev set, one remedy is to collect more training data that better _____ those difficult examples.
Match each part of a traffic-sign recognition scenario to its role in a data mismatch diagnosis.
Order the steps for deciding whether to collect training data that better matches difficult dev examples.
Using Targeted Data to Reduce Distribution Mismatch
Fix a training-dev distribution gap
What training data change helps with a data mismatch problem?
Learn After
Overfitting from Reused Background Noise
Synthetic Data Can Still Cause Narrow Memorization
What is the main concern when synthetic examples are easy to tell apart from real data?
Representativeness of Synthetic Noise Samples
Synthetic Data Should Reflect the Real Task
Match each synthetic-data case to whether it is representative.
Order the steps for checking whether generated examples are truly representative.
Why synthetic training examples must reflect real variation
Check a Synthesized Audio Data Plan for Hidden Artifacts
How to check whether synthetic training data looks unrepresentative
Which synthetic-data strategy is least likely to create a representativeness problem?
True or False: Synthetic data built from only a small set of source examples can be unrepresentative.