Lottery Ticket Hypothesis
Definition
The Lottery Ticket Hypothesis states that large randomly initialized networks can contain smaller subnetworks that are effectively trainable with suitable initialization.
"Winning tickets" are typically found through pruning experiments. The hypothesis inspired research on sparsity and pruning but is not a general guarantee for every model.
Simply explained
Like finding a small particularly favorable substructure early in a large network.
Why it matters in practice
Relevant for model compression and fundamental research.
Example from business
An experiment prunes a network iteratively and trains subnetworks with reset initializations.
Limits and pitfalls
- The hypothesis is not a general recipe for success in production. Searching for suitable subnetworks also incurs computational cost.
Origin & history
Frankle and Carbin introduced the Lottery Ticket Hypothesis in 2018/2019.
Use in mid-sized companies
- Possible application in SMEs: Mainly classify as a research concept for sparsity and pruning.
Frequently asked questions
Is a winning ticket known before training?
Usually not. Does LTH mean large models are unnecessary? Not directly.
Sources
- The Lottery Ticket Hypothesis · Frankle & Carbin (2019), The Lottery Ticket Hypothesis
Sources checked on 10.10.2026.
Editorially reviewed on: