Research map / Architectures and efficiency / CNN design
Accelerating Convolutional Neural Networks via Activation Map Compression
Check 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 |
|---|---|---|---|
| Add an activation-map sparsity-inducing prior during fine-tuning | LeNet-5 · MNIST | Top-1 accuracy: 98.45% -> 98.48% (+0.03%) | Helped |
| Add an activation-map sparsity-inducing prior during fine-tuning | MobileNet-V1 · CIFAR-10 | Top-1 accuracy: 89.17% -> 89.71% (+0.54%) | Helped |
| Add an activation-map sparsity-inducing prior during fine-tuning | Inception-V3 · ImageNet | Top-1 accuracy: 75.76% -> 76.14% (+0.38%) | Helped |
| Increase sparsity pressure to the Sparse_v2 setting (ablation) | Inception-V3 · ImageNet | Top-1 accuracy: 75.76% -> 68.94% (-6.82%) | Hurt |
| Increase sparsity pressure to the Sparse_v2 setting (ablation) | ResNet-18 · ImageNet | Top-1 accuracy: 69.64% -> 68.62% (-1.02%) | Hurt |
| Increase sparsity pressure to the Sparse_v2 setting (ablation) | ResNet-34 · ImageNet | Top-1 accuracy: 73.26% -> 67.73% (-5.53%) | Hurt |
| Quantize activation maps to uint16 after sparsification | LeNet-5 Sparse · MNIST | Top-1 accuracy: 98.48% -> 98.48% (+0.03%) | Helped |
| Quantize activation maps to uint8 after sparsification (ablation) | LeNet-5 Sparse · MNIST | Compression gain: 8.43× (3.16×) -> 11.16× (2.79×) | Helped |
| Use SEG rather than EG for entropy coding (ablation) | Sparse Inception-V3 activation maps · ImageNet | Compression gain: 4.10× (2.05×) -> 5.80× (2.90×) | Helped |
| Sparsify activation maps before uint16 quantization and SEG coding | LeNet-5 · MNIST | Compression gain: 3.40× (1.70×) -> 6.76× (3.38×) | Helped |
Closest papers by meaning
- Inducing and Exploiting Activation Sparsity for Fast Inference on Deep Neural Networks (ICML 2020 · similarity 0.90)
- Activation Map Compression through Tensor Decomposition for Deep Learning (NeurIPS 2024 · similarity 0.88)
- Wavelet Feature Maps Compression for Image-to-Image CNNs (NeurIPS 2022 · similarity 0.88)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.87)
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.86)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · 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.