Research map / Architectures and efficiency / Pruning and efficient tuning
Accelerating CNN Training by Pruning Activation Gradients
Xucheng Ye, Pengcheng Dai, Junyu Luo, Xin Guo, Yingjie Qi, Jianlei Yang, Yiran Chen
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 |
|---|---|---|---|
| raise pruning rate to p=99% (ablation) | stochastic gradient pruning, AlexNet · ImageNet | training accuracy (%): 56.38 -> 39.58 | Hurt |
| raise pruning rate to p=99% (ablation) | stochastic gradient pruning, ResNet-152 · CIFAR-10 | training accuracy (%): 95.70 -> 93.84 | Hurt |
| raise pruning rate to p=99% (ablation) | stochastic gradient pruning, ResNet-152 · CIFAR-100 | training accuracy (%): 79.25 -> 76.40 | Hurt |
| raise pruning rate to p=99% (ablation) | stochastic gradient pruning, ResNet-101 · CIFAR-100 | training accuracy (%): 79.70 -> 77.13 | Hurt |
| apply stochastic gradient pruning at p=70% | ResNet-50 · CIFAR-100 | training accuracy (%): 77.74 -> 78.83 | Improved |
| apply stochastic gradient pruning at p=70% | ResNet-18 · ImageNet | training accuracy (%): 68.73 -> 69.02 | No clear effect |
| apply stochastic gradient pruning at p=70% | ResNet-34 · ImageNet | training accuracy (%): 72.93 -> 72.92 | No clear effect |
| apply stochastic gradient pruning at p=99% | AlexNet · CIFAR-10 | training accuracy (%): 90.50 -> 89.66 | Hurt |
| apply stochastic gradient pruning at p=80% | AlexNet · CIFAR-100 | training accuracy (%): 67.61 -> 68.13 | Improved |
| apply stochastic gradient pruning at p=99% | ResNet-18 · CIFAR-10 | non-zero density: 1 -> 0.16 | Improved |
| use distribution-based threshold with stochastic pruning instead of MSBP | ResNet-18 · CIFAR-10 | non-zero density: 0.4 -> 0.16 | Improved |
| use distribution-based threshold with stochastic pruning instead of MSBP | ResNet-18 · CIFAR-10 | accuracy (%): 94.92 -> 95.18 | Improved |
Closest papers by meaning
- Inducing and Exploiting Activation Sparsity for Fast Inference on Deep Neural Networks (ICML 2020 · similarity 0.92)
- Faster CNNs with Direct Sparse Convolutions and Guided Pruning (ICLR 2017 · similarity 0.90)
- Growing Efficient Deep Networks by Structured Continuous Sparsification (ICLR 2021 · similarity 0.89)
- Accelerate CNN via Recursive Bayesian Pruning (ICCV 2019 · similarity 0.89)
- Neural Pruning via Growing Regularization (ICLR 2021 · similarity 0.89)
- Sparsity in Deep Learning: Pruning and growth for efficient inference and training in neural networks (JMLR 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.
