Research map / Architectures and efficiency / RNNs and normalization
Adaptive Activation Thresholding: Dynamic Routing Type Behavior for Interpretability in Convolutional Neural Networks
Yiyou Sun, Sathya N. Ravi, Vikas Singh
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 |
|---|---|---|---|
| replace final layer with Dynamic-K (K=25) activation | ResNet18 · Places365 val | top-1 accuracy (%): 53.69 -> 53.87 | No clear effect |
| replace final layer with Dynamic-K (K=5) | ResNet18 · Places365 val | top-1 accuracy (%): 53.69 -> 53.60 | No clear effect |
| use Dynamic-K with K=1 instead of larger K (ablation) | ResNet18 · Places365 val | top-1 accuracy (%): 53.69 -> 27.83 | Hurt |
| replace final layer with Dynamic-K (K=25) | ResNet50 · Places365 val | top-1 accuracy (%): 54.74 -> 55.00 | No clear effect |
| replace final layer with Dynamic-K (K=5) | ResNet18 · CIFAR-10 | accuracy (%): 95.10 -> 95.12 | No clear effect |
| replace final layer with Dynamic-K (K=25) | ResNet50 · CIFAR-10 | accuracy (%): 95.62 -> 95.58 | No clear effect |
| use small K (Dynamic-K) instead of l1 regularization for sparsity | ResNet18 · CIFAR-10 final-layer sparsity | sparsity | Improved |
| decrease K in Dynamic-K activation (ablation) | ResNet18 · Places365 NetDissect IoU | interpretability (IoU) | Improved |
| increase K to 100 or more (ablation) | ResNet18 · Places365 final-layer sparsity | sparse connection percentage | Hurt |
Closest papers by meaning
- Deciding How to Decide: Dynamic Routing in Artificial Neural Networks (ICML 2017 · similarity 0.87)
- Differentiable Dynamic Wirings for Neural Networks (ICCV 2021 · similarity 0.86)
- DISCOVER: Making Vision Networks Interpretable via Competition and Dissection (NeurIPS 2023 · similarity 0.86)
- Latency-Aware Unified Dynamic Networks for Efficient Image Recognition (TPAMI 2024 · similarity 0.86)
- SkipNet: Learning Dynamic Routing in Convolutional Networks (ECCV 2018 · similarity 0.86)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 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.
