Research map / Architectures and efficiency / Neural architecture search
SuperTickets: Drawing Task-Agnostic Lottery Tickets from Supernets via Jointly Architecture Searching and Parameter Pruning
Haoran You, Baopu Li, Zhanyi Sun, Xu Ouyang, Yingyan Lin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add joint search-and-prune two-in-one training instead of search-then-prune (ablation) | S+P (Mag.) · Cityscapes 90% sparsity | mIoU (%): 42.12 -> 55.84 | Improved |
| add progressive pruning (ablation) | two-in-one training · Cityscapes 90% sparsity | mIoU (%): 55.84 -> 63.89 | Improved |
| add progressive pruning (ablation) | two-in-one training · COCO keypoint 80% sparsity | AP (%): 58.38 -> 60.14 | Improved |
| reactivate pruned weights' gradients right after pruning (IR-P) (ablation) | two-in-one + PP · Cityscapes 90% sparsity | mIoU (%): 63.89 -> 45.73 | Hurt |
| reactivate pruned weights' gradients during search (IR-S) (ablation) | two-in-one + PP · Cityscapes 90% sparsity | mIoU (%): 63.89 -> 66.61 | Improved |
| reactivate pruned weights' gradients during search (IR-S) (ablation) | two-in-one + PP · COCO keypoint 80% sparsity | AP (%): 60.14 -> 61.02 | Improved |
| add 50 epochs of retraining to SuperTickets (ablation) | SuperTickets · Cityscapes 90% sparsity | mIoU (%): 66.61 -> 67.17 | Improved |
| add 50 epochs of retraining to SuperTickets (ablation) | SuperTickets · COCO keypoint 80% sparsity | AP (%): 61.02 -> 61.48 | Improved |
| retrain S+P (Mag.) baseline | S+P (Mag.) · Cityscapes transfer 90% sparsity | mIoU (%): 42.12 -> 60.76 | Improved |
| transfer SuperTickets from ADE20K to Cityscapes | SuperTickets · Cityscapes 90% sparsity | mIoU (%): 60.76 -> 62.91 | Improved |
| transfer ImageNet classification tickets to ADE20K segmentation vs Cityscapes tickets | SuperTickets · ADE20K 90% sparsity | mIoU (%): 20.83 -> 22.42 | Improved |
| use SuperTickets instead of S+P (Mag.) at 50% pruning | SuperTickets · ImageNet | top-1 accuracy (%): 72.8 -> 74.2 | Improved |
Closest papers by meaning
- The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (ICLR 2019 · similarity 0.91)
- Exploring Lottery Ticket Hypothesis in Spiking Neural Networks (ECCV 2022 · similarity 0.89)
- Deconstructing Lottery Tickets: Zeros, Signs, and the Supermask (NeurIPS 2019 · similarity 0.89)
- Efficient Lottery Ticket Finding: Less Data is More (ICML 2021 · similarity 0.89)
- Lottery Jackpots Exist in Pre-Trained Models (TPAMI 2023 · similarity 0.89)
- Multi-Prize Lottery Ticket Hypothesis: Finding Accurate Binary Neural Networks by Pruning A Randomly Weighted Network (ICLR 2021 · similarity 0.89)
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.
