Research map / Optimization / Implicit bias and generalization
Understanding Impacts of High-Order Loss Approximations and Features in Deep Learning Interpretation
Sahil Singla, Eric Wallace, Shi Feng, Soheil Feizi
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 |
|---|---|---|---|
| Add the Hessian term to first-order interpretation (ablation) | CAFO on ImageNet ILSVRC-2012 with ResNet-50 · ImageNet ILSVRC-2012 validation, high-confidence predictions | interpretation similarity | No clear effect |
| Add the Hessian term to first-order interpretation (ablation) | CAFO on ImageNet ILSVRC-2012 with ResNet-50 · ImageNet ILSVRC-2012 validation, low-confidence predictions | interpretation difference | Mixed |
| Use group-feature sparsity regularization instead of isolated feature importance (ablation) | CAFO and CASO · ImageNet ILSVRC-2012 | visual coherence | Improved |
| Use group-feature sparsity regularization instead of isolated feature importance (ablation) | CAFO and CASO · ImageNet ILSVRC-2012 | saliency-map noise | Improved |
| Use ordinary gradient descent rather than proximal gradient descent with the L1 penalty (ablation) | CAFO and CASO with positive L1 regularization · ImageNet interpretation optimization | convergence quality | Hurt |
| Increase the L1 regularization parameter until the interpretation becomes all zeros (ablation) | CAFO and CASO · ImageNet ILSVRC-2012 | saliency-map sparsity | Mixed |
| Multiply importance values by raw input pixels (ablation) | saliency-map visualization methods · Image interpretation | visual artifacts | Hurt |
Closest papers by meaning
- Investigating the Overlooked Hessian Structure: From CNNs to LLMs (ICML 2025 · similarity 0.87)
- Analytic Insights into Structure and Rank of Neural Network Hessian Maps (NeurIPS 2021 · similarity 0.87)
- A Deeper Look at the Hessian Eigenspectrum of Deep Neural Networks and its Applications to Regularization (AAAI 2021 · similarity 0.87)
- The Hessian perspective into the Nature of Convolutional Neural Networks (ICML 2023 · similarity 0.87)
- Why Do Better Loss Functions Lead to Less Transferable Features? (NeurIPS 2021 · similarity 0.86)
- Structured Gradient-based Interpretations via Norm-Regularized Adversarial Training (CVPR 2024 · 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.
