Research map / Learning theory and algorithms / Optimal transport
AvAtar: Learning to Align via Active Optimal Transport
Qi Yu, Ruizhong Qiu, Zhichen Zeng, My T. Thai, huan liu, Hanghang Tong
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 |
|---|---|---|---|
| weight each pair's query impact by the transport plan row as a posterior (ablation) | AvAtar-L2 · Douban (PARROT) | MRR: 0.669 -> 0.837 | Improved |
| weight each pair's query impact by the transport plan row as a posterior (ablation) | AvAtar-Consist · Phone-Email (PARROT) | MRR: 0.506 -> 0.630 | Improved |
| use sparse matrix operations in conjugate gradient instead of dense (ablation) | AvAtar-L2 · Douban (PARROT) | MRR: 0.839 -> 0.837 | No clear effect |
| use sparse matrix operations in conjugate gradient instead of dense (ablation) | AvAtar-Consist · ACM-DBLP-A (PARROT) | query time (s): 1010 -> 118 | Improved |
| use gradient-propagated entropy utility instead of applying entropy directly to T | active query for GOT · image-text retrieval and grounding | Recall@1: up to 2.6% in Recall@1 | Improved |
| use AvAtar-L2/Consist instead of best baseline query strategies | OT-based network alignment · Phone-Email, ACM-DBLP-P, Douban, ACM-DBLP-A | MRR: up to 2.8% in MRR | Improved |
| use network-specific GibbsMatchings/TopMatchings instead of generic Entropy | OT-based network alignment · Network alignment with PARROT and JOENA | MRR: JOENA Phone-Email round 10: 0.778 -> 0.661 (GibbsMatchings) | Hurt |
| use consistency utility instead of L2 utility | AvAtar · plain networks (Phone-Email, ACM-DBLP-P) | MRR: up to 0.6% improvement on plain networks; L2 better on attributed | Mixed |
| use entropy utility instead of L2 utility | AvAtar · image-text retrieval and grounding | Recall@1: ImageNet-C GOT-W round 10: 0.505 -> 0.509 | Mixed |
| vary query batch size nb (ablation) | AvAtar-L2 · Phone-Email and Douban (PARROT) | MRR | No clear effect |
| increase query budget k (ablation) | AvAtar · all tasks | MRR / Recall@1: Phone-Email PARROT: 0.539 -> 0.629 (round 5 -> 10) | Improved |
| use AvAtar instead of TopMatchings/GibbsMatchings for efficiency | active network alignment · Phone-Email and Douban (PARROT) | query time and MRR: up to 25x speed-up; up to 8.1% MRR | Improved |
Closest papers by meaning
- Hierarchical Multi-Marginal Optimal Transport for Network Alignment (AAAI 2024 · similarity 0.86)
- Efficient Transferable Optimal Transport via Min-Sliced Transport Plans (NeurIPS 2026 · similarity 0.85)
- Optimizing Distributional Geometry Alignment with Optimal Transport for Generative Dataset Distillation (NeurIPS 2025 · similarity 0.84)
- Integrating Efficient Optimal Transport and Functional Maps For Unsupervised Shape Correspondence Learning (CVPR 2024 · similarity 0.84)
- Aligning Embeddings and Geometric Random Graphs: Informational Results and Computational Approaches for the Procrustes-Wasserstein Problem (NeurIPS 2024 · similarity 0.84)
- Value Gradient Guidance for Flow Matching Alignment (NeurIPS 2025 · 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.
