Research map / Architectures and efficiency / Pruning and efficient tuning
Lottery Jackpots Exist in Pre-Trained Models
Yuxin Zhang, Mingbao Lin, Yunshan Zhong, Fei Chao, Rongrong Ji
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 magnitude-based mask initialization (ablation) | SR-popup · CIFAR-10 ResNet-32 90% | top-1 accuracy: 94.01 -> 94.30 | Improved |
| add short restriction on weight swapping (ablation) | SR-popup · CIFAR-10 ResNet-32 90% | top-1 accuracy: 93.69 -> 94.30 | Improved |
| select swap pairs with largest expected loss drops instead of inverse (ablation) | SR-popup · CIFAR-10 ResNet-32 90% | top-1 accuracy: 93.13 -> 94.30 | Improved |
| select swap pairs with largest expected loss drops instead of random (ablation) | SR-popup · CIFAR-10 ResNet-32 90% | top-1 accuracy: 93.84 -> 94.30 | Improved |
| use declining swap schedule instead of constant (ablation) | SR-popup · CIFAR-10 ResNet-32 | top-1 accuracy: 93.13 -> 93.51 | Improved |
| use declining swap schedule instead of inverse schedule (ablation) | SR-popup · CIFAR-10 ResNet-32 | top-1 accuracy: 93.44 -> 93.51 | Improved |
| replace Edge-popup with SR-popup (30 search epochs) | Pre-trained ResNet-50 · ImageNet 90% | top-1 accuracy: 70.20 -> 72.61 | Improved |
| replace Edge-popup with SR-popup (60 search epochs) | Pre-trained ResNet-50 · ImageNet 90% | top-1 accuracy: 70.89 -> 73.04 | Improved |
| increase searching from 10 to 60 epochs | Jackpot · CIFAR-10 ResNet-32 90% | top-1 accuracy: 94.03 -> 94.43 | Improved |
| increase searching from 100 to 1000 epochs | Jackpot · CIFAR-10 ResNet-32 90% | top-1 accuracy: 94.41 -> 94.43 | No clear effect |
| fine-tune weights after searching mask | Lottery jackpot · CIFAR-10 ResNet-32 | top-1 accuracy | No clear effect |
| prune VGGNet-19 with jackpot instead of dense | Jackpot 30 epochs · CIFAR-100 90% | top-1 accuracy: 74.16 -> 74.63 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SNIP — SNIP: SINGLE-SHOT NETWORK PRUNING BASED ON CONNECTION SENSITIVITY (ICLR 2019) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks (ICLR 2019 · similarity 0.91)
- Deconstructing Lottery Tickets: Zeros, Signs, and the Supermask (NeurIPS 2019 · similarity 0.91)
- Lottery Ticket Preserves Weight Correlation: Is It Desirable or Not? (ICML 2021 · similarity 0.91)
- Winning the Lottery with Continuous Sparsification (NeurIPS 2020 · similarity 0.91)
- Efficient Lottery Ticket Finding: Less Data is More (ICML 2021 · similarity 0.91)
- Find A Winning Sign: Sign Is All We Need to Win the Lottery (ICLR 2025 · 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.
