Upsampling
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
Two techniques share the name: adding minority-class examples to balance a training dataset, and increasing the resolution of an image or signal.
What is Upsampling?
Upsampling means one of two things: duplicating or synthesizing minority-class examples to balance a dataset, or increasing the resolution of an image or the sampling rate of a signal.
Context decides which sense is meant. A data scientist fixing a fraud dataset and an audio engineer converting a recording both “upsample,” and the operations have nothing in common beyond adding data points.
Types of Upsampling
Upsampling for class balance (also called oversampling) adds minority-class examples to imbalanced data so the model sees rare cases often enough to learn them. Random upsampling duplicates existing minority examples; SMOTE generates synthetic data by interpolating between them; ADASYN concentrates the new points where the class is hardest to separate.
Upsampling for resolution inserts new samples or pixels into a signal, raising an image’s dimensions or an audio stream’s sampling rate, with interpolation filling the gaps. Neural networks use it too: the decoder of an autoencoder and the generator of a GAN contain upsampling layers that grow small feature maps back into full-sized images.
How Upsampling for Class Balance Works
The mechanics carry two warnings. Duplicated examples tempt the model to memorize them, a route to overfitting, which is why synthetic methods like SMOTE exist.
And upsampling must happen after the train/test split, never before. Balance the whole dataset first and copies of the same example land on both sides of the split, a form of data leakage that inflates test scores on examples the model has effectively already seen.
Example of Upsampling
A bank trains a fraud classifier on 100,000 transactions of which 1,000 are fraudulent. Trained as-is, the model learns that predicting “legitimate” every time is 99% accurate, and catches nothing.
The team splits off a test set first, then applies SMOTE to the training portion, generating synthetic fraud cases until the classes are roughly even. Retrained, the model finally learns what fraud looks like, and the untouched test set gives an honest measure of it.
The other sense, for contrast: a photo-editing tool upsampling a 100x100 thumbnail to 400x400 invents the missing pixels by interpolation. Same word, different world.
Related AI terms: Imbalanced Data · Synthetic Data · Data Leakage · Sampling Rate · Data Augmentation
Did you like the Upsampling gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About