Omniglot

E899066

benchmark dataset few-shot learning benchmark machine learning dataset one-shot learning benchmark

Omniglot is a widely used benchmark dataset of handwritten characters from numerous alphabets, designed to evaluate one-shot and few-shot learning in machine learning research.

Jump to: Statements Referenced by

Statements (51)

Predicate	Object
instanceOf	benchmark dataset ⓘ few-shot learning benchmark ⓘ machine learning dataset ⓘ one-shot learning benchmark ⓘ
backgroundSetUsedFor	meta-training ⓘ
benchmarkFor	Bayesian program learning NERFINISHED ⓘ matching networks ⓘ memory-augmented neural networks ⓘ meta-learning algorithms ⓘ model-agnostic meta-learning ⓘ neural Turing machines NERFINISHED ⓘ prototypical networks NERFINISHED ⓘ
collectionMethod	handwriting on a digital tablet ⓘ online crowd-sourcing ⓘ
comparedTo	ImageNet in terms of role as a benchmark NERFINISHED ⓘ
contains	characters from constructed alphabets ⓘ characters from multiple alphabets ⓘ characters from real-world writing systems ⓘ
creator	Brenden M. Lake NERFINISHED ⓘ Joshua B. Tenenbaum NERFINISHED ⓘ Ruslan Salakhutdinov NERFINISHED ⓘ
dataType	handwritten characters ⓘ
describedIn	Human-level concept learning through probabilistic program induction NERFINISHED ⓘ
domain	handwritten character recognition ⓘ
eachCharacterHas	20 instances ⓘ
eachClassDrawnBy	multiple different people ⓘ
evaluationSetUsedFor	meta-testing ⓘ
hostedAt	author-maintained project website ⓘ
imageModality	grayscale images ⓘ
imageSize	105x105 pixels ⓘ
includes	50 different alphabets ⓘ alphabets from diverse language families ⓘ invented alphabets created for the dataset ⓘ
inspiredBy	the need for a character-level ImageNet for one-shot learning NERFINISHED ⓘ
license	freely available for research use ⓘ
numberOfCharactersPerAlphabet	approximately 20 to 40 GENERATED ⓘ
numberOfClasses	1623 ⓘ
numberOfExamplesPerClass	20 GENERATED ⓘ
numberOfImages	32460 ⓘ
publicationVenue	Proceedings of the 27th Annual Conference of the Cognitive Science Society NERFINISHED ⓘ
publicationYear	2015 ⓘ
taskSupported	few-shot classification ⓘ generative modeling ⓘ one-shot classification ⓘ sequence prediction of pen strokes ⓘ
typicalSplit	background set and evaluation set ⓘ
usedFor	evaluating few-shot learning algorithms ⓘ evaluating one-shot learning algorithms ⓘ learning to learn experiments ⓘ meta-learning research ⓘ testing rapid concept learning from few examples ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

Matching Networks for One Shot Learning → datasetUsed → Omniglot ⓘ