Research map / Architectures and efficiency / RNNs and normalization
Quantification and Analysis of Layer-wise and Pixel-wise Information Discarding
Haotian Ma, Hao Zhang, Fan Zhou, Yinqing Zhang, Quanshi Zhang
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 |
|---|---|---|---|
| Use CID rather than CAM or Grad-CAM for localization maps | ResNet-50 on CUB200-2011 · CUB200-2011 weakly-supervised localization | localization accuracy: CAM 0.367; Grad-CAM 0.355; CID 0.493 | Improved |
| Use CID rather than Grad-CAM or LRP for localization maps | VGG-16 on CUB200-2011 · CUB200-2011 weakly-supervised localization | localization accuracy: Grad-CAM 0.507; LRP 0.489; CID 0.578 | Improved |
| Compress VGG-16 to 38.0% parameters (ablation) | VGG-16 on CUB200-2011 · CUB200-2011 VGG-16 compression | adversarial robustness ||ε||₂: ΔCID -0.003; adversarial robustness 0.00281 | Mixed |
| Compress VGG-16 to 12.8% parameters (ablation) | VGG-16 on CUB200-2011 · CUB200-2011 VGG-16 compression | adversarial robustness ||ε||₂: ΔCID -0.005; adversarial robustness 0.00254 | Hurt |
| Compress VGG-16 to 8.0% parameters (ablation) | VGG-16 on CUB200-2011 · CUB200-2011 VGG-16 compression | adversarial robustness ||ε||₂: ΔCID -0.033; adversarial robustness 0.00109 | Hurt |
| Compress the network (ablation) | VGG-16 on CUB200-2011 · CUB200-2011 compression | CID | Hurt |
| Compress the network (ablation) | VGG-16 on CUB200-2011 · CUB200-2011 compression | reconstruction capacity | No clear effect |
| Compress the network (ablation) | VGG-16 on CUB200-2011 · CUB200-2011 compression | concentration | No clear effect |
| Train AlexNet with knowledge distillation (ablation) | AlexNet on CUB200-2011 · CUB200-2011 classification | information discarding (CID) | Hurt |
| Increase network depth | DNNs for image classification and reconstruction · Image classification and image reconstruction | CID | Mixed |
| Use adversarial training (ablation) | AlexNet on CUB-200 2011 · CUB-200 2011 classification | CID | Improved |
| Revise a VGG convolutional layer to use four 7×7×512 filters (ablation) | VGG-16/VGG-19 · CUB200-2011 bird versus ImageNet random-image classification | CID | Hurt |
Closest papers by meaning
- Information Bottleneck Analysis of Deep Neural Networks via Lossy Compression (ICLR 2024 · similarity 0.87)
- Estimating Information Flow in Deep Neural Networks (ICML 2019 · similarity 0.86)
- Information Bottleneck: Exact Analysis of (Quantized) Neural Networks (ICLR 2022 · similarity 0.85)
- From Alexnet to Transformers: Measuring the Non-linearity of Deep Neural Networks with Affine Optimal Transport (CVPR 2025 · similarity 0.84)
- Understanding and Leveraging the Learning Phases of Neural Networks (AAAI 2024 · similarity 0.84)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.83)
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.
