Research map / Learning theory and algorithms / Submodular and game algorithms
Neural Estimation of Submodular Functions with Applications to Differentiable Subset Selection
Abir De, Soumen Chakrabarti
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 |
|---|---|---|---|
| Use FlexSubNet instead of the strongest baseline | Synthetic set-function estimation · Log, LogDet, Facility location, monotone graph cut, Log×sqrt, Log×LogDet, non-monotone graph cut | RMSE: Log: 0.060 $\pm$ 0.001 -> 0.015 $\pm$ 0.000; LogDet: 0.029 $\pm$ 0.000 -> 0.013 $\pm$ 0.000; Facility location: 0.063 $\ | Improved |
| Use FlexSubNet instead of DSF | Amazon subset selection · Gear, Bath, Health, Diaper, Toys, Bedding, Feeding, Apparel, Media | MJC: Gear: 0.099 -> 0.101; Bath: 0.087 -> 0.091; Health: 0.142 -> 0.153; Diaper: 0.115 -> 0.134; Toys: 0.150 -> 0.157; Beddin | Improved |
| Use FlexSubNet instead of DSF | Amazon subset selection · Gear, Bath, Health, Diaper, Toys, Bedding, Feeding, Apparel, Media | NDCG@10: Gear: 0.538 -> 0.539; Bath: 0.500 -> 0.520; Health: 0.549 -> 0.597; Diaper: 0.546 -> 0.562; Toys: 0.577 -> 0.591; Beddin | Improved |
| Decouple the double integral into independent integrals (ablation) | FlexSubNet · Log, LogDet, Facility location | RMSE: Log: 0.078 $\pm$ 0.001 -> 0.015 $\pm$ 0.000; LogDet: 0.073 $\pm$ 0.001 -> 0.013 $\pm$ 0.000; Facility location: 0.078 $\ | Improved |
| Use one recursion step instead of the full two-step model (ablation) | FlexSubNet · Log, LogDet, Facility location | RMSE: Log: 0.015 $\pm$ 0.000 -> 0.089 $\pm$ 0.001; LogDet: 0.013 $\pm$ 0.000 -> 0.075 $\pm$ 0.001; Facility location: 0.022 $\ | Hurt |
| Train end-to-end through double integrals instead of decoupling (ablation) | FlexSubNet · Log, LogDet, Facility location | RMSE: Log: 0.015 $\pm$ 0.000 -> 0.078 $\pm$ 0.001; LogDet: 0.013 $\pm$ 0.000 -> 0.073 $\pm$ 0.001; Facility location: 0.022 $\ | Hurt |
| Add an additional component to the SubMix model (ablation) | SubMix · Synthetic set-function estimation | RMSE | No clear effect |
| Use the initial SubMix form with added 1 terms inside the logarithms (ablation) | SubMix · Synthetic set-function estimation | RMSE | Hurt |
Closest papers by meaning
- Non-Monotone DR-Submodular Function Maximization (AAAI 2017 · similarity 0.87)
- Joint Continuous and Discrete Model Selection via Submodularity (JMLR 2022 · similarity 0.87)
- Deep Submodular Peripteral Networks (NeurIPS 2024 · similarity 0.86)
- Enhancing Neural Subset Selection: Integrating Background Information into Set Representations (ICLR 2024 · similarity 0.86)
- Online Two-Stage Submodular Maximization (NeurIPS 2025 · similarity 0.85)
- Globally Convergent Variational Inference (NeurIPS 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.
