Before Active Learning: Prerequisites, Building Blocks, and the Hard Limits of Query Strategies
Active learning lets a model pick which examples to label instead of sampling at random — but sampling bias and cold-start can make it lose to random.
Training Data Quality & Curation
Strategies for building high-quality training datasets including cleaning, labeling, augmentation, and deduplication.
This theme is curated by our AI council — see how it works.
What topics does this domain cover?
6 topicsEach topic below is a key concept in this domain. Pick any for the full picture: foundations, implementation, what's changing, and risks to consider.
Active Learning →
Active learning is a machine learning strategy where the model itself picks the most informative unlabeled examples for …
6 articles
Data Augmentation →
Data augmentation expands a training dataset by creating new examples from existing ones—rotating or cropping images, …
6 articles
Data Deduplication →
Data deduplication finds and removes duplicate or near-duplicate examples from a training dataset before a model learns …
6 articles
Data Labeling and Annotation →
Data labeling and annotation is the process of attaching ground-truth labels to raw data — text, images, audio, or video …
6 articles
Data Preprocessing →
Data preprocessing is the work of cleaning, normalizing, and transforming raw data into a form a machine learning model …
6 articles
Training Data Quality →
Training data quality measures how clean, consistent, and correct the examples used to train a machine learning model …
6 articles
Four perspectives on this domain
MONA's articles build your mental model — how things work, why they work that way, and what intuition to develop.
Updated Jun 7, 2026
Concepts covered
Exact, Fuzzy, and Semantic Deduplication: The Components and Prerequisites of a Dedup Pipeline
Data deduplication runs in three tiers: exact (hashing), fuzzy (MinHash+LSH), and semantic (embeddings). SemDeDup removed ~50% of web data with minimal loss.
False Positives, Lost Diversity, and the Technical Limits of Deduplicating Training Data
Data deduplication measures surface overlap, not meaning, so it deletes rare examples as false positives and misses reworded copies. Here are the limits.
Uncertainty Sampling Explained: Entropy, Margin, and Least-Confidence Query Strategies
Uncertainty sampling is an active-learning strategy that labels the data a model is least confident about — via entropy, margin, or least-confidence scores.
What Is Active Learning and How Models Pick the Most Informative Samples to Label
Active learning lets a model query only the most informative unlabeled samples to label, hitting target accuracy with far fewer labels than random sampling.
What Is Data Deduplication and How MinHash LSH Detects Near-Duplicate Training Samples
Data deduplication removes near-duplicate training samples using MinHash LSH. Lee et al. found dedup cuts verbatim memorization roughly 10x.
Before You Preprocess: Data Types, Distributions, and Train-Test Splits You Need to Understand First
Split data into train and test sets before preprocessing to prevent data leakage. Fitting scalers on the full dataset inflates accuracy and fails in production.
Data Leakage, Lost Information, and the Technical Limits of Preprocessing Pipelines
Data leakage occurs when information unavailable at prediction time enters training, inflating validation accuracy while production performance collapses.
What Is Data Preprocessing and How Cleaning, Scaling, and Encoding Turn Raw Data into Training Sets
Data preprocessing cleans, scales, and encodes raw data into model-ready features. Fitting transformers before the train-test split causes data leakage.
Geometric Transforms, Mixup, and Back-Translation: How Core Augmentation Methods Work
Data augmentation transforms existing examples — flips, mixup blends, CutMix patches, back-translation — to teach models invariance, not add raw data.
Inter-Annotator Agreement, Annotation Guidelines, and the Building Blocks of a Labeling Project
Inter-annotator agreement measures label quality beyond chance. Cohen's kappa corrects raw match rates, exposing unreliable labels that 90% agreement hides.
Label Noise, Annotator Bias, and the Technical Limits of Human Data Annotation
Label noise averages an estimated 3.4% across major ML test sets, distorting supervised model accuracy and even flipping benchmark leaderboard rankings.
What Is Data Augmentation and How Transforming Samples Expands Training Data
Data augmentation expands training data by transforming existing samples—rotations, mixup, masking—to reduce overfitting without collecting anything new.
What Is Data Labeling and Annotation, and How Ground-Truth Labels Train Supervised Models
Data labeling assigns ground-truth labels to raw data so supervised models learn a mapping. Label noise propagates into model errors geometrically.
When Data Augmentation Helps and When It Hurts: Distribution Shift and Label Corruption
Data augmentation helps until synthetic samples drift from real data or break the input-label mapping, creating distribution shift and label corruption.
Label Noise, Class Imbalance, and Distribution Shift: What to Know Before Fixing Training Data
Label noise, class imbalance, and distribution shift degrade models more than architecture choices. Understand all three before curating training data.
What Is Training Data Quality and How It Determines Model Performance
Training data quality is the systematic engineering of label correctness, deduplication, and provenance — it sets the ceiling on what any model can learn.
Why Perfectly Clean Data Is Impossible: The Technical Limits of Data Curation at Scale
Cleaning training data at scale hits hard limits: label errors average 3.4% across top ML datasets, and automated cleaners misfire on half their flags.
