Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Hierarchical Variational Memory for Few-shot Learning Across Domains
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 hierarchical variational prototypes rather than prototypes from only the last layer (ablation) | Variational prototype model · Cross-domain few-shot benchmarks, 5-way 5-shot | accuracy (%): 80.72 -> 83.75; 73.21 -> 74.29; 40.53 -> 41.21; 25.03 -> 26.15 | Helped |
| Replace flat variational memory with hierarchical memory (ablation) | Variational memory · Cross-domain few-shot benchmarks, 5-way 5-shot | accuracy (%): 79.12 -> 87.65; 72.21 -> 74.88; 38.97 -> 42.05; 22.15 -> 27.15 | Helped |
| Learn prototype-level weights instead of using bagging (ablation) | Hierarchical variational memory · Cross-domain few-shot benchmarks, 5-way 5-shot | accuracy (%): 85.43 -> 87.65; 74.02 -> 74.88; 40.39 -> 42.05; 25.98 -> 27.15 | Helped |
| Add hierarchical formulation for prototypes (ablation) | Few-shot learner with Conv-4 · miniImagenet, 5-way 1-shot | accuracy (%): 54.73 -> 56.72 | Helped |
| Add hierarchical formulation for prototypes (ablation) | Few-shot learner with Conv-4 · miniImagenet, 5-way 5-shot | accuracy (%): 68.01 -> 70.11 | Helped |
| Add hierarchical formulation for prototypes (ablation) | Few-shot learner with Conv-4 · tieredImagenet, 5-way 1-shot | accuracy (%): 56.88± 1.71 -> 57.89± 0.81 | Helped |
| Add hierarchical formulation for prototypes (ablation) | Few-shot learner with Conv-4 · tieredImagenet, 5-way 5-shot | accuracy (%): 74.65± 0.81 -> 76.01± 0.60 | Helped |
| Combine prototypes from all levels rather than using only the last level (ablation) | Hierarchical variational memory, Conv-4 · miniImagenet, 5-way 1-shot | accuracy (%): 56.72 -> 57.04 | Helped |
| Combine prototypes from all levels rather than using only the last level (ablation) | Hierarchical variational memory, Conv-4 · miniImagenet, 5-way 5-shot | accuracy (%): 70.11 -> 72.65 | Helped |
| Combine prototypes from all levels rather than using only the last level (ablation) | Hierarchical variational memory, Conv-4 · tieredImagenet, 5-way 1-shot | accuracy (%): 57.89± 0.81 -> 59.01 | Helped |
| Combine prototypes from all levels rather than using only the last level (ablation) | Hierarchical variational memory, Conv-4 · tieredImagenet, 5-way 5-shot | accuracy (%): 76.01± 0.60 -> 77.76 | Helped |
| Increase backbone capacity (ablation) | Hierarchical variational memory · miniImagenet and tieredImagenet, 5-way 1-shot and 5-shot | accuracy (%) | Helped |
Closest papers by meaning
- Learning to Learn Variational Semantic Memory (NeurIPS 2020 · similarity 0.93)
- Variational Memory Addressing in Generative Models (NeurIPS 2017 · similarity 0.89)
- MetaModulation: Learning Variational Feature Hierarchies for Few-Shot Learning with Fewer Tasks (ICML 2023 · similarity 0.87)
- Dual Memory Networks: A Versatile Adaptation Approach for Vision-Language Models (CVPR 2024 · similarity 0.86)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.85)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.85)
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.