Research map / Architectures and efficiency / Pruning and efficient tuning
Lottery Tickets on a Data Diet: Finding Initializations with Sparse Trainable Networks
Mansheej Paul, Brett W Larsen, Surya Ganguli, Jonathan Frankle, Gintare Karolina Dziugaite
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 |
|---|---|---|---|
| increase IMP rewind step from 0 to t* | IMP sparse networks · CIFAR-10/CIFAR-100/CINIC-10 (ResNet-20/32/56) | final test accuracy at high sparsity | Improved |
| pre-train on small random subset for t*/2 instead of all data for t*/2 | IMP initialization · CIFAR-10/CIFAR-100/CINIC-10 (ResNet-20/32/56) | final test accuracy | Improved |
| pre-train on easiest (lowest EL2N) examples for t*/2 | IMP initialization · CIFAR-10/CIFAR-100/CINIC-10 (ResNet-20/32/56) | final test accuracy | Improved |
| pre-train on hardest (highest EL2N) examples for t*/2 | IMP initialization · CIFAR-10/CIFAR-100/CINIC-10 (ResNet-20/32/56) | final test accuracy | Hurt |
| add 10% randomized labels during pre-training: easiest-data subset vs random subset or all data | IMP initialization · CIFAR-10 | final test accuracy | Improved |
| increase randomized labels to 50% during pre-training with all data or random subset | IMP initialization · CIFAR-10 | final test accuracy | Hurt |
| pre-train on lowest-train-loss-barrier subset (ablation) | IMP initialization · CIFAR-100/CINIC-10 | final test accuracy | Mixed |
| pre-train on easy data (hypothesis: acts as gradient clipping) (ablation) | dense network pre-training · CIFAR-10/CIFAR-100/CINIC-10 (ResNet-20/32/56) | minibatch gradient norm | Hurt |
| use mean per-example train loss barrier as predictor instead of pre-train accuracy | dense initialization · CIFAR-10/CIFAR-100/CINIC-10 (ResNet-20/32/56) | correlation with final test accuracy | Improved |
Closest papers by meaning
- Unmasking the Lottery Ticket Hypothesis: What's Encoded in a Winning Ticket's Mask? (ICLR 2023 · similarity 0.91)
- No Free Prune: Information-Theoretic Barriers to Pruning at Initialization (ICML 2024 · similarity 0.90)
- Pruning at Initialization – A Sketching Perspective (TPAMI 2025 · similarity 0.90)
- Towards Data-Agnostic Pruning At Initialization: What Makes a Good Sparse Mask? (NeurIPS 2023 · similarity 0.89)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.87)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.87)
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.
