Lottery Ticket Hypothesis
A neural network hypothesis proposed by Frankle and Carlin in 2019, stating that large randomly initialized networks contain smaller subnetworks that, when trained in isolation, can achieve performance comparable to the full network. The hypothesis provides theoretical grounding for the compressibility of large models.
Timeline (last 90 days)
The lottery ticket hypothesis was proposed by Frankle and Carlin in 2019, pointing out that within a randomly initialized large neural network there exists a smaller subnetwork that, when trained in isolation, can achieve performance comparable to or even better than the original network.
Lottery ticket subnetworks often exhibit low-rank or sparse characteristics, which closely align with the low-rank structure exploited by latent factor decomposition.
All Facts (2)
The lottery ticket hypothesis was proposed by Frankle and Carlin in 2019, pointing out that within a randomly initialized large neural network there exists a smaller subnetwork that, when trained in isolation, can achieve performance comparable to or even better than the original network.
50%UnverifiedLottery ticket subnetworks often exhibit low-rank or sparse characteristics, which closely align with the low-rank structure exploited by latent factor decomposition.
50%