Research map / Label-efficient and robust learning / Self-supervised and contrastive
Function Contrastive Learning of Transferable Meta-Representations
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 |
|---|---|---|---|
| Train the encoder with function-level contrastive learning and freeze it before training downstream decoders | FCRL representations · Sinusoid test functions, few-shot regression | MSE: 0.310±0.05 (NP), 0.265±0.03 (CNP) -> 0.172±0.04 (FCRL) at 5-shot; 0.218±0.02 (NP), 0.149±0.02 (CNP) -> 0.100±0.02 (FCRL) | Helped |
| Train the encoder with function-level contrastive learning and freeze it before training downstream decoders | FCRL representations · Sinusoid test functions, few-shot parameter identification | MSE: 0.0087±0.0007 (NP), 0.0096±0.0007 (CNP) -> 0.0078±0.0004 at 5-shot; 0.0037±0.0005 (NP), 0.0049±0.0011 (CNP) -> 0.0032±0. | Helped |
| Use FCRL representations instead of jointly trained CNP representations | MNIST function image completion · MNIST functions, 50-shot FSIC | MSE: 0.0477±\num{0.0006} -> 0.0481±\num{0.0001} | No clear effect |
| Use FCRL representations instead of jointly trained CNP representations | MNIST function image completion · MNIST functions, 200-shot FSIC | MSE: 0.0347±\num{0.0011} -> 0.0355±\num{0.0001} | No clear effect |
| Increase the number of partial observations J (ablation) | FCRL on MNIST2D · MNIST2D validation, FSCC | accuracy | Helped |
| Use a nonlinear critic rather than alternative critics (ablation) | FCRL on MNIST2D · MNIST2D validation, FSCC | accuracy | Helped |
| Vary the temperature parameter τ (ablation) | FCRL on MPI3D · MPI3D downstream classification | accuracy | No clear effect |
| Use a nonlinear critic rather than alternative critics (ablation) | FCRL on MPI3D · MPI3D downstream classification | accuracy | Helped |
Closest papers by meaning
- MetaFun: Meta-Learning with Iterative Functional Updates (ICML 2020 · similarity 0.90)
- Provable Meta-Learning of Linear Representations (ICML 2021 · similarity 0.89)
- Conditional Meta-Learning of Linear Representations (NeurIPS 2022 · similarity 0.89)
- Learning to Learn with Contrastive Meta-Objective (NeurIPS 2025 · similarity 0.88)
- Provable Meta-Learning with Low-Rank Adaptations (NeurIPS 2025 · similarity 0.88)
- First-order ANIL provably learns representations despite overparametrisation (ICLR 2024 · 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.