Research map / Architectures and efficiency / CNN design
Thanks for Nothing: Predicting Zero-Valued Activations with Lightweight Convolutional Neural Networks
Gil Shomron, Ron Banner, Moran Shkolnik, Uri Weiser
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 |
|---|---|---|---|
| fine-tune the model after deploying ZAP (ablation) | ZAP · CIFAR-100 AlexNet | MAC reduction at about 1% top-1 drop: 32.4% MAC reduction with 0.7% degradation -> 40.3% with 0.8% degradation | Improved |
| fine-tune the model after ZAP (ablation) | ZAP · ILSVRC-2012 AlexNet (vs Shomron) | top-1 accuracy degradation: 1.28 -> 0.78 | Improved |
| fine-tune the model after ZAP (ablation) | ZAP · ILSVRC-2012 ResNet-18 (vs Shomron) | top-1 accuracy degradation: 2.96 -> 1.86 | Improved |
| fine-tune the model after ZAP (ablation) | ZAP · ILSVRC-2012 VGG-16 (vs Shomron) | top-1 accuracy degradation: 1.71 -> 0.31 | Improved |
| use masks with larger computed ratio alpha (mask A vs D) (ablation) | ZAP · CIFAR-100 AlexNet | top-1 accuracy degradation | Mixed |
| raise the prediction threshold (ablation) | ZAP · CIFAR-100 AlexNet | misprediction of non-zero activations | Hurt |
| replace Figurnov et al. interpolation with ZAP | ZAP · ILSVRC-2012 AlexNet | top-5 accuracy degradation: 8.50 -> 3.04 | Improved |
| replace Figurnov et al. with ZAP | ZAP · ILSVRC-2012 VGG-16 | top-5 accuracy degradation: 15.6 -> 7.02 | Improved |
| replace Shomron and Weiser with ZAP | ZAP · ILSVRC-2012 ResNet-18 | top-1 accuracy degradation: 11.0 -> 2.96 | Improved |
| replace Kim et al. with ZAP | ZAP · ILSVRC-2012 VGG-16 | top-5 accuracy degradation: 0.26 -> 0.33 | No clear effect |
| replace Dong et al. with ZAP | ZAP · ILSVRC-2012 ResNet-18 | top-5 accuracy degradation fine-tuned: 2.3 -> 4.66 | Hurt |
Closest papers by meaning
- Beyond the Limits: Overcoming Negative Correlation of Activation-Based Training-Free NAS (ICCV 2025 · similarity 0.85)
- Zero Stability Well Predicts Performance of Convolutional Neural Networks (AAAI 2022 · similarity 0.85)
- L-SWAG: Layer-Sample Wise Activation with Gradients Information for Zero-Shot NAS on Vision Transformers (CVPR 2025 · similarity 0.85)
- Linear CNNs Discover the Statistical Structure of the Dataset Using Only the Most Dominant Frequencies (ICML 2023 · similarity 0.84)
- Multi-Bias Non-linear Activation in Deep Neural Networks (ICML 2016 · similarity 0.83)
- Do We Always Need the Simplicity Bias? Looking for Optimal Inductive Biases in the Wild (CVPR 2025 · similarity 0.82)
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.
