Research map / Learning theory and algorithms / Prediction and decision losses
Streaming Weak Submodularity: Interpreting Neural Networks on the Fly
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 Streak instead of LIME for selecting explanatory superpixels | Inception V3 flower-classifier explanations · Inception V3 flower-classifier explanations, N=30, k=5 | runtime: 10 times faster | Helped |
| Use Streak for superpixel selection | Inception V3 flower-classifier explanations · Inception V3 flower-classifier explanations | explanation quality | No clear effect |
| Use Streak instead of LocalSearch | Sparse logistic regression with pairwise features · Phishing dataset | log likelihood | No clear effect |
| Use Streak instead of LocalSearch | Sparse logistic regression with pairwise features · Phishing dataset | generalization accuracy | No clear effect |
| Use RandomSubset instead of LocalSearch | Sparse logistic regression with pairwise features · Phishing dataset | log likelihood | Hurt |
| Use RandomSubset instead of LocalSearch | Sparse logistic regression with pairwise features · Phishing dataset | generalization accuracy | Hurt |
| Set ε=0.1 rather than ε=0.75 (ablation) | Streak for sparse logistic regression · Phishing dataset, k=80 | runtime relative to LocalSearch at k=40: 5% -> 70% | Mixed |
| Replace LIME feature selection with Streak and retain LIME's final regression (ablation) | LIME + Streak · Inception V3 flower-classifier explanations | explanation quality | Mixed |
| Replace LIME feature selection with Streak and retain LIME's final regression (ablation) | LIME + Streak · Inception V3 flower-classifier explanations | runtime | No clear effect |
| Vary ε to trade precision for speed (ablation) | Streak for sparse logistic regression · Phishing dataset, k=80 | log likelihood versus runtime | Mixed |
Closest papers by meaning
- Do Less, Get More: Streaming Submodular Maximization with Subsampling (NeurIPS 2018 · similarity 0.87)
- Streaming Robust Submodular Maximization: A Partitioned Thresholding Approach (NeurIPS 2017 · similarity 0.87)
- Weakly Submodular Maximization Beyond Cardinality Constraints: Does Randomization Help Greedy? (ICML 2018 · similarity 0.86)
- Beyond 1/2-Approximation for Submodular Maximization on Massive Data Streams (ICML 2018 · similarity 0.86)
- Algorithm Development in Neural Networks: Insights from the Streaming Parity Task (ICML 2025 · similarity 0.85)
- Consistent Submodular Maximization (ICML 2024 · similarity 0.84)
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.