Research map / Architectures and efficiency / Pruning and efficient tuning
The Elastic Lottery Ticket Hypothesis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Transfer a winning ticket across depths within the same model family using layer replication or dropping and reordering | ResNet tickets · ImageNet | Top-1 accuracy: 68.17% (ResNet-18); 73.37% (ResNet-34) | Helped |
| Transfer a ticket between architectures using ETTs rather than training a dynamic sparse model independently | ResNet-18/26/34 · ImageNet | Top-1 accuracy: 68.17% vs 69.59%; 72.94% vs 72.66%; 73.37% vs 73.88% | Mixed |
| Transfer a ResNet-20 ticket to ResNet-32 with ETTs | ResNet-32 · CIFAR-10, 90% sparsity | Accuracy drop: -1.41% -> -0.96% | Helped |
| Transfer a ResNet-32 ticket to ResNet-56 with ETTs | ResNet-56 · CIFAR-10, 95% sparsity | Accuracy drop: -1.77% -> -1.66% | Helped |
| Transfer a ResNet-14 ticket to ResNet-20 with ETTs | ResNet-20 · CIFAR-10, 95% sparsity | Accuracy drop: -4.47% -> -4.48% | No clear effect |
| Transfer tickets across FC-layer configurations | VGG-13 and MLP · CIFAR-10 and MNIST | Accuracy | Helped |
| Replicate more unique residual blocks rather than only the earliest block (ablation) | ResNet-32 ticket stretched to ResNet-56 · CIFAR-10 | Accuracy | Helped |
| Replicate earlier rather than later residual blocks (ablation) | ResNet-32 ticket stretched to ResNet-56 · CIFAR-10, high sparsity | Accuracy | Helped |
| Drop the first four residual blocks when squeezing the ticket (ablation) | ResNet-56 ticket squeezed to ResNet-32 · CIFAR-10 | Accuracy | Hurt |
| Choose consecutive versus non-consecutive blocks to drop (ablation) | ResNet-56 ticket squeezed to ResNet-32 · CIFAR-10 | Accuracy | No clear effect |
| Replicate source blocks in interpolation rather than appending order (ablation) | ResNet tickets · CIFAR-10 | Accuracy | No clear effect |
| Use ETT mask/weight transformations instead of copying existing source masks and using SNIP for added blocks (ablation) | ResNet-32 ticket stretched to ResNet-56 · CIFAR-10, high sparsity | Accuracy | Helped |
Closest papers by meaning
- Coarsening the Granularity: Towards Structurally Sparse Lottery Tickets (ICML 2022 · similarity 0.94)
- Dual Lottery Ticket Hypothesis (ICLR 2022 · similarity 0.94)
- Find A Winning Sign: Sign Is All We Need to Win the Lottery (ICLR 2025 · similarity 0.93)
- The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (ICLR 2019 · similarity 0.93)
- Weisfeiler and Leman Go Gambling: Why Expressive Lottery Tickets Win (ICML 2025 · similarity 0.90)
- On the Sparsity of the Strong Lottery Ticket Hypothesis (NeurIPS 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.