Research map / Architectures and efficiency / RNNs and normalization
Learning Important Features Through Propagating Activation Differences
Read the paperCheck 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 |
|---|---|---|---|
| use RevealCancel rule instead of other backprop attribution methods | DeepLIFT · MNIST digit conversion (pixel erasing) | increase in log-odds of target vs original class | Helped |
| use Rescale instead of RevealCancel on various layers (ablation) | DeepLIFT · MNIST digit conversion | log-odds increase | Hurt |
| integrated gradients with 10 intervals instead of 5 (ablation) | Integrated gradients · MNIST digit conversion | log-odds increase: 5 -> 10 intervals | No clear effect |
| integrated gradients versus gradient*input | attribution baselines · MNIST digit conversion | log-odds increase | No clear effect |
| Guided Backprop discarding negative gradients | Guided Backprop · MNIST digit conversion | log-odds increase | Hurt |
| RevealCancel only on fully connected layer, Rescale on conv layers (ablation) | DeepLIFT · TAL-GATA DNA simulation | noise in importance scores | Helped |
| use RevealCancel on fully connected layer instead of Rescale only (ablation) | DeepLIFT · TAL-GATA DNA simulation | missed TAL1 relevance on Task 0 | Helped |
| Guided Backprop x input versus DeepLIFT RevealCancel | Guided Backprop x input · TAL-GATA DNA simulation | false positives for TAL1 on task 1 | Hurt |
Closest papers by meaning
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks (NeurIPS 2025 · similarity 0.85)
- Expand-and-Cluster: Parameter Recovery of Neural Networks (ICML 2024 · similarity 0.84)
- Synthesizing the preferred inputs for neurons in neural networks via deep generator networks (NeurIPS 2016 · similarity 0.84)
- Neural Collapse in Cumulative Link Models for Ordinal Regression: An Analysis with Unconstrained Feature Model (NeurIPS 2025 · similarity 0.84)
- Network Inversion of Convolutional Neural Nets (Student Abstract) (AAAI 2025 · similarity 0.83)
- Improving Deep Learning Speed and Performance Through Synaptic Neural Balance (AAAI 2025 · 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.