Research map / Architectures and efficiency / RNNs and normalization
Improving Deep Learning Interpretability by Saliency Guided Training
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 |
|---|---|---|---|
| Apply saliency-guided training | CNN on MNIST · MNIST | accuracy-drop AUC: 3360.4 -> 1817.6 | Helped |
| Fine-tune with saliency-guided training rather than train with it from scratch (ablation) | CNN on MNIST · MNIST | accuracy-drop AUC: 1817.6 -> 2258.8 | Hurt |
| Apply saliency-guided training | BiLSTM with Integrated Gradients · Movie Reviews | comprehensiveness AOPC: 0.265 -> 0.306 | Helped |
| Apply saliency-guided training | BiLSTM with Integrated Gradients · Movie Reviews | sufficiency AOPC: 0.054 -> 0.002 | Helped |
| Apply saliency-guided training | BiLSTM with SmoothGrad · FEVER | sufficiency AOPC: 0.006 -> 0.006 | No clear effect |
| Apply saliency-guided training | TCN with Integrated Gradients · Synthetic time series (10-dataset mean) | Diff(AUP): 0.233 -> 0.291 | Helped |
| Apply saliency-guided training | TCN with Integrated Gradients · Synthetic time series (10-dataset mean) | Diff(AUR): 0.191 -> 0.210 | Helped |
| Apply saliency-guided training | Transformer saliency explanations · Synthetic time series | precision | No clear effect |
| Apply saliency-guided training | Transformer saliency explanations · Synthetic time series | recall | Helped |
| Replace low-saliency language tokens with the previous high-saliency word (ablation) | Saliency-guided language training · Movie Reviews, FEVER, and e-SNLI | interpretability | Helped |
Closest papers by meaning
- Benchmarking Deep Learning Interpretability in Time Series Predictions (NeurIPS 2020 · similarity 0.89)
- Shallow and Deep Convolutional Networks for Saliency Prediction (CVPR 2016 · similarity 0.88)
- Understanding and Visualizing Deep Visual Saliency Models (CVPR 2019 · similarity 0.87)
- Learning to Promote Saliency Detectors (CVPR 2018 · similarity 0.87)
- Progressive Attention Guided Recurrent Network for Salient Object Detection (CVPR 2018 · similarity 0.87)
- Saliency-Guided Image Translation (CVPR 2021 · similarity 0.86)
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.