Research map / Architectures and efficiency / Pruning and efficient tuning
PAC-Net: A Model Pruning Approach to Inductive Transfer Learning
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 |
|---|---|---|---|
| Use PAC-Net instead of fine-tuning | Inductive transfer on modified Friedman #1 · Modified Friedman #1, d=1, 10 target samples | RMSE: 10.8 -> 6.7 | Helped |
| Use PAC-Net instead of L^2-SP-Fisher | ResNet-18 transfer on CelebA · Arched Eyebrows to Eyeglasses, average | accuracy: 66.9 -> 69.5 | Helped |
| Use PAC-Net instead of fine-tuning | ResNet-18 transfer on CelebA · Eyeglasses to Arched Eyebrows, average | accuracy: 69.7 -> 75.5 | Helped |
| Use PAC-Net instead of L^2-SP-Fisher | Neural ODE transfer on Duffing equation · Linear oscillator, average | RMSE: 0.5 -> 0.07 | Helped |
| Use PAC-Net instead of fine-tuning | Neural ODE transfer on Duffing equation · Non-linear oscillator, average | RMSE: 0.60 -> 0.42 | Helped |
| Use PAC-Net instead of fine-tuning | PINN transfer on diffusion equation · Diffusion equation, target task, average | RMS error: 0.074 -> 0.030 | Helped |
| Use PAC-Net instead of training on target data only | RTT transfer on semiconductor dataset · Semiconductor dataset, 40 target samples | score: 0.799 -> 0.907 | Helped |
| Add source-weight allocation retraining after pruning (ablation) | PAC-Net · Modified Friedman #1, d=1, average | RMSE: 7.5 -> 4.4 | Helped |
| Add L^2 regularization on pruned weights (ablation) | PAC-Net · Modified Friedman #1, d=1, average | RMSE: 5.7 -> 4.4 | Helped |
| Initialize pruned weights at zero rather than randomly (ablation) | PAC-Net · Modified Friedman #1, d=1, average | RMSE: 5.6 -> 4.4 | Helped |
| Keep source weights fixed rather than updating all weights with soft L^2-SP (ablation) | PAC-Net · Modified Friedman #1, d=1, average | RMSE: 6.3 -> 4.4 | Helped |
| Keep source weights fixed rather than updating all weights with L^2-SP-Fisher (ablation) | PAC-Net · Modified Friedman #1, d=1, average | RMSE: 9.1 -> 4.4 | Helped |
Closest papers by meaning
- Selectivity Drives Productivity: Efficient Dataset Pruning for Enhanced Transfer Learning (NeurIPS 2023 · similarity 0.87)
- Sparse Deep Transfer Learning for Convolutional Neural Network (AAAI 2017 · similarity 0.87)
- Explicit Inductive Bias for Transfer Learning with Convolutional Networks (ICML 2018 · similarity 0.87)
- Pruning Convolutional Neural Networks for Resource Efficient Inference (ICLR 2017 · similarity 0.87)
- How Well Do Sparse ImageNet Models Transfer? (CVPR 2022 · similarity 0.86)
- MULTIFLOW: Shifting Towards Task-Agnostic Vision-Language Pruning (CVPR 2024 · similarity 0.85)
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.