Research map / Bandits and online learning / Online convex optimization and games
Optimal-er Auctions through Attention
Dmitry Ivanov, Iskander Safiulin, Igor Filippov, Ksenia Balabaeva
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 |
|---|---|---|---|
| Replace RegretNet's fully connected architecture with attention-based RegretFormer | auction mechanism learning · 2x2, 2x3, 2x5, and 3x10; regret budget 10^-3 | revenue: 0.889 -> 0.908; 1.317 -> 1.416; 2.339 -> 2.453; 5.59 -> 6.121 | Improved |
| Replace RegretNet's fully connected architecture with attention-based RegretFormer | auction mechanism learning · 1x2, regret budget 10^-3 | revenue: 0.572 -> 0.571 | Hurt |
| Replace RegretNet or EquivariantNet with RegretFormer | auction mechanism learning · 1x2 through 3x10; regret budget 10^-4 | revenue: 0.551 -> 0.556; 0.825 -> 0.861; 1.249 -> 1.327; 2.121 -> 2.339; 5.02 -> 5.745 | Improved |
| Add positional encoding to RegretFormer (ablation) | RegretFormer · asymmetric one bidder, two items | revenue: 9.967 -> 10.056 | Improved |
| Use RegretFormer rather than RegretNet in multi-setting training | auction mechanism learning · multi-setting average | revenue: 2.573 -> 2.703 | Improved |
| Use RegretFormer rather than RegretNet for out-of-multi-setting generalization | auction mechanism learning · out-of-multi-setting average | regret: 0.0662 -> 0.00478 | Improved |
| Tighten the regret budget from 10^-3 to 10^-4 | RegretFormer · 1x2 | revenue: 0.571 -> 0.556 | Hurt |
| Tighten the regret budget from 10^-3 to 10^-4 | RegretFormer · 2x2 | revenue: 0.908 -> 0.861 | Hurt |
| Tighten the regret budget from 10^-3 to 10^-4 | RegretFormer · 2x3 | revenue: 1.416 -> 1.327 | Hurt |
| Tighten the regret budget from 10^-3 to 10^-4 | RegretFormer · 2x5 | revenue: 2.453 -> 2.339 | Hurt |
| Tighten the regret budget from 10^-3 to 10^-4 | RegretFormer · 3x10 | revenue: 6.121 -> 5.745 | Hurt |
| Use EquivariantNet rather than RegretFormer | auction mechanism learning · multi-setting | regret: 0.00391 -> 0.00989 | Hurt |
Closest papers by meaning
- PreferenceNet: Encoding Human Preferences in Auction Design with Deep Learning (NeurIPS 2021 · similarity 0.92)
- Comparing Uniform Price and Discriminatory Multi-Unit Auctions through Regret Minimization (NeurIPS 2025 · similarity 0.88)
- AuctionNet: A Novel Benchmark for Decision-Making in Large-Scale Games (NeurIPS 2024 · similarity 0.87)
- Auctionformer: A Unified Deep Learning Algorithm for Solving Equilibrium Strategies in Auction Games (ICML 2024 · similarity 0.87)
- Learning Optimal Auctions with Correlated Value Distributions (AAAI 2025 · similarity 0.86)
- BundleFlow: Deep Menus for Combinatorial Auctions by Diffusion-Based Optimization (NeurIPS 2025 · 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.
