A data scientist explains: "Our synthetic data has high fidelity but low utility for the downstream classification task."
What does this statement mean?
Fidelity vs. Utility — they can diverge:
Metric type
What it measures
How measured
Fidelity
Statistical similarity to real data
JSD, KS test, correlation matrix comparison
Utility
Performance of models trained on synthetic data
TSTR accuracy, F1 score, AUC
Why the divergence? High-fidelity distributions may still lack the decision-relevant signal needed for the task. Key vocabulary: TSTR gap, downstream utility, evaluation benchmark, task-specific utility.
3 / 11
A privacy auditor asks: "Can a classifier distinguish synthetic from real data?" and sets up a binary classification experiment.
What does this experiment measure, and what result indicates high-quality synthetic data?
Classifier-based fidelity evaluation:
Classifier accuracy
Interpretation
≈ 50%
Cannot distinguish → very high fidelity synthetic data
60–70%
Some detectable differences; moderate fidelity
> 80%
Clear distributional gap; low fidelity
This is also called a discriminative score or detection test. Key vocabulary: discriminative score, detection test, synthetic vs. real classifier, fidelity metric, distributional gap.
4 / 11
A team evaluates synthetic data quality using a utility score computed from multiple downstream tasks.
Which of the following best describes what a "utility score" represents in synthetic data evaluation?
Utility score in synthetic data evaluation:
Component
How it contributes to the utility score
TSTR accuracy
Classification performance on real test data
Regression MSE
Error of models trained on synthetic vs. real data
Query accuracy
Accuracy of aggregate queries on synthetic vs. real data
Alex: 'The synthetic dataset's utility score is only 75% – that's significantly lower than our real data. We need to investigate why.' What does Alex likely mean by 'utility score'?
Utility score in this context refers to the practical usefulness of the synthetic data for training machine learning models. It's not about raw statistical similarity (option 1), but rather how well it translates into model performance on real tasks. A low utility score suggests that despite perhaps resembling the real data, the synthetic data isn't effectively helping models learn useful patterns (option 3). The size of the dataset itself doesn't directly determine utility (option 4).
7 / 11
Sarah: 'I've run a basic classification model on our synthetic customer data, and the accuracy is only 60%. This isn't good enough for production.' What does Sarah likely mean when she mentions the 'accuracy' of the model?
Accuracy in machine learning specifically refers to the proportion of correctly classified instances. Sarah's concern is that the model isn't performing well enough on the synthetic data – a low accuracy suggests significant errors in classification. It's crucial to understand that accuracy doesn't directly reflect complexity or resource usage.
8 / 11
During a code review, Ben comments on a pull request: 'The synthetic data generator is producing very noisy labels. We're seeing significant disagreement between the generated features and their corresponding labels. This will severely impact model training.' What does Ben's comment primarily highlight about the quality of the synthetic data?
Ben's comment focuses on 'noisy labels,' which indicates low utility. 'Utility' in synthetic data evaluation refers to how well the data can be used for training a model; noisy labels directly undermine this. High fidelity implies accurate representation of real-world characteristics, but that doesn't guarantee useful labels.
9 / 11
Maria, a data privacy engineer, is assessing a synthetic dataset. She runs a differential privacy experiment and observes that the output distribution of the generated data closely matches the real data distribution. What does this suggest about the quality of the synthetic data?
When differential privacy experiments show a close match between synthetic and real distributions, it's a strong indicator of 'leakage.' This means information from the original dataset is still present in the synthetic data. While utility might be high (the data *is* useful), it comes at the cost of diminished privacy.
10 / 11
David is presenting a report on synthetic data evaluation. He states: 'We're using a utility score based on the performance of our fraud detection model trained on the synthetic data – it's currently 82%.'. What does 'utility score' represent in this context?
In synthetic data evaluation, the 'utility score' directly reflects the performance of a downstream task when trained on the synthetic data. It's a measure of how useful the generated data is for achieving that specific goal – in this case, fraud detection accuracy. It's distinct from diversity or computational cost.
11 / 11
During a Slack discussion about synthetic data generation, Chloe says: 'We're generating data with a high degree of fidelity to the real distribution. However, our model trained on this data performs poorly when deployed in production.' What is Chloe most likely referring to?
While 'fidelity' describes how closely the synthetic data mirrors the real distribution, it doesn't guarantee 'utility.' High fidelity means the generated data *looks* similar to reality, but if that similarity doesn't translate into good performance on a downstream task (like in production), the data has low utility. The other options are potential issues, but Chloe's primary concern is the mismatch between fidelity and utility.
What will I learn from the "Synthetic Data Evaluation Vocabulary Exercises" exercise?
Practice English vocabulary for synthetic data evaluation: fidelity metrics, downstream task performance, the classifier distinction test, and utility scores.
Is this exercise free to use?
Yes. Every exercise on CoderSlingo, including this one, is free to use with no account, sign-up, or paywall required.
How many questions are in this exercise?
This set contains 11 multiple-choice questions, each with a detailed explanation shown after you answer.
Do I need to create an account to track my progress?
No account is required. Your progress bar and score reset each time you reload the page, but you can retry the exercise as many times as you like.
Who is this Synthetic Data Vocabulary exercise for?
This exercise is built for IT professionals and non-native English speakers who need to read, write, and discuss synthetic data vocabulary topics confidently at work.
What happens if I answer a question incorrectly?
You will see the correct answer highlighted along with a detailed explanation of why it is correct -- so every wrong answer becomes a learning moment, not just a lost point.
Can I retry this exercise?
Yes -- click "Try again" on the results screen at any time to reset your score and go through all the questions again.
How long does this exercise take to complete?
Most learners finish all 11 questions in under 10 minutes, since each question is answered by clicking a single option.
Where can I find more Synthetic Data Vocabulary exercises?
See the full Synthetic Data Vocabulary exercises hub for more vocabulary drills on this topic.
Is this exercise mobile-friendly?
Yes -- the exercise works on any device with a modern browser, including phones and tablets, with no app download required.