Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Episodic Multi-Task Learning with Heterogeneous Neural Processes
Jiayi Shen, Xiantong Zhen, Cheems Wang, Marcel Worring
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 |
|---|---|---|---|
| remove both global latent z and local latent parameters w (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 67.09 -> 76.29 | Improved |
| add global latent representation z (vs. w only) (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 73.29 -> 76.29 | Improved |
| add local latent parameters w (vs. z only) (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 71.11 -> 76.29 | Improved |
| replace probabilistic transformer inference with MLP (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 76.29 -> 73.30 | Hurt |
| remove learnable tokens from transformer inference (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 76.29 -> 75.25 | Hurt |
| use deterministic instead of probabilistic inference modules (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 76.29 -> 70.61 | Hurt |
| remove learnable tokens, deterministic setting (ablation) | HNPs deterministic · Office-Home 4-task 5-way 1-shot | average accuracy: 70.22 -> 70.61 | Improved |
| generate w by concatenating z with context features (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 76.29 -> 69.74 | Hurt |
| generate w by adding z to context features (ablation) | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy: 76.29 -> 72.87 | Hurt |
| increase shots from 1 to 20 (HNPs vs TNP-D gap) | HNPs vs TNP-D · Office-Home 4-task 5-way | average accuracy: 1-shot 76.29 vs 65.49; 20-shot 81.56 vs 81.12 | Mixed |
| increase number of classes from 20 to 40 | HNPs vs TNP-D · DomainNet 6-task 1-shot | average accuracy: 40-way 29.31 vs 21.64 | Mixed |
| vary number of Monte Carlo samples from 1 to 30 | HNPs · Office-Home 4-task 5-way 1-shot | average accuracy | No clear effect |
Closest papers by meaning
- Efficient Heterogeneous Meta-Learning via Channel Shuffling Modulation (ICLR 2024 · similarity 0.88)
- Multi-Task Processes (ICLR 2022 · similarity 0.86)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.86)
- A Hierarchical Bayesian Model for Few-Shot Meta Learning (ICLR 2024 · similarity 0.85)
- Learning to Learn with Contrastive Meta-Objective (NeurIPS 2025 · similarity 0.85)
- Amalgamating Multi-Task Models with Heterogeneous Architectures (AAAI 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.
