Research map / Label-efficient and robust learning / Recognition and long-tail
Task2Vec: Task Embedding for Meta-Learning
Alessandro Achille, Michael Lam, Rahul Tewari, Avinash Ravichandran, Subhransu Maji, Charless C. Fowlkes, Stefano Soatto, Pietro Perona
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 asymmetric distance instead of symmetric Task2Vec (ablation) | Task2Vec expert selection · iNat + CUB | relative error increase over optimal (%): +42.54% -> +9.97% | Improved |
| use asymmetric distance instead of symmetric Task2Vec (ablation) | Task2Vec expert selection · Mixed | relative error increase over optimal (%): +40.30% -> +29.23% | Improved |
| add learned model bias (Model2Vec) to asymmetric Task2Vec (ablation) | expert selection · iNat + CUB | relative error increase over optimal (%): +9.97% -> +6.81% | Improved |
| add learned model bias (Model2Vec) to asymmetric Task2Vec (ablation) | expert selection · Mixed | relative error increase over optimal (%): +29.23% -> +27.81% | Improved |
| select expert with Asymmetric Task2Vec instead of generic ImageNet expert | feature extractor choice · iNat + CUB | relative error increase over optimal (%): +30.18% -> +9.97% | Improved |
| select expert with Asymmetric Task2Vec instead of generic ImageNet expert | feature extractor choice · Mixed | relative error increase over optimal (%): +75.73% -> +29.23% | Improved |
| use ResNet-13 instead of VGG-13 as probe network (ablation) | Task2Vec · iNat + CUB (all tasks) | relative error increase over optimal (%): +38.03% -> +9.97% | Improved |
| use DenseNet-121 instead of VGG-13 as probe network (ablation) | Task2Vec · iNat + CUB (all tasks) | relative error increase over optimal (%): +38.03% -> +10.63% | Improved |
| use task embedding instead of domain embedding (mean feature activations) (ablation) | task representation · iMaterialist tasks | relative error increase over optimal (%) | Improved |
| vary number of training samples for the task | Task2Vec model selection · Mixed subset (data efficiency) | relative error increase over optimal (%) | No clear effect |
Later papers that compared with this method
3 later accepted papers report Task2Vec as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- ModelLens: Finding the Best for Your Task from Myriads of Models (NeurIPS 2026 · better on 3 of 3 shared results)
- Learning to Group Auxiliary Datasets for Molecule (NeurIPS 2023 · better on 3 of 3 shared results)
- An Information-Geometric Distance on the Space of Tasks (ICML 2021 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MATE: Plugging in Model Awareness to Task Embedding for Meta Learning (NeurIPS 2020 · similarity 0.89)
- Revisiting Unsupervised Meta-Learning via the Characteristics of Few-Shot Tasks (TPAMI 2022 · similarity 0.89)
- Task2Box: Box Embeddings for Modeling Asymmetric Task Relationships (CVPR 2024 · similarity 0.86)
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 2024 · similarity 0.86)
- Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · similarity 0.86)
- Learning Task-Agnostic Representations through Multi-Teacher Distillation (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.
