MENTAL MODEL · M3215

Lottery Ticket Hypothesis

TechnicalHigh supportNeuroscience

Included

Version 1.0.0 · Updated 2026-07-31

CORE DEFINITION

A hypothesis proposed by Jonathan Frankle and Michael Carbin, suggesting that within a randomly initialized neural network, there exists a sparse subnetwork (the "winning ticket") that, when trained in isolation, can achieve performance comparable to the full network. This subnetwork has a special initial weight configuration from the start.

SCAFFOLDING EFFECT

psychology

Reduce cognitive load

Explanation of network sparsity. It provides a new perspective for understanding the training dynamics and generalization ability of neural networks, revealing the "lucky" initial configurations that exist in networks, and has important implications for model compression and efficient training.

anchor

Anchor fast decisions

The hypothesis posits that within a randomly initialized neural network, there exists a sparse subnetwork (the "winning ticket") that, when trained in isolation with its initial weights, can achieve performance comparable to the full network, indicating that large models contain discoverable "efficient substructures".

MINIMUM ACTION

In progress 0/4

Practice this model in one real situation:

Randomly initialize a network and train it to obtain a mask
prune by weight magnitude to retain the subnetwork
reset the subnetwork to its original initial weights and train it in isolation
verify whether it can match the original performance.

Check to track your progress (stored locally)

Learning progress0%

account_treeGenealogyexpand_more

menu_bookReferencesexpand_more

Source support: Explicit

link
en.wikipedia.orghttps://en.wikipedia.org/wiki/Lottery_ticket_hypothesisZH · Explicit
verified

RELATED MODELS