Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Transformer Neural Processes: Uncertainty-Aware Meta Learning Via Sequence Modeling
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 TNP-A rather than the strongest NP baseline | Meta regression · RBF | log-likelihood: 0.82 \pm 0.01 -> 1.63 \pm 0.00 | Helped |
| Use TNP-A rather than the strongest NP baseline | Meta regression · Matérn 5/2 | log-likelihood: 0.66 \pm 0.00 -> 1.21 \pm 0.00 | Helped |
| Use TNP-A rather than the strongest NP baseline | Meta regression · Periodic | log-likelihood: -0.96 \pm 0.02 -> -2.26 \pm 0.17 | Hurt |
| Use TNP-A rather than the strongest NP baseline | Image completion · CelebA | log-likelihood: 3.09 \pm 0.00 -> 5.82 \pm 0.01 | Helped |
| Use TNP-A rather than the strongest NP baseline | Image completion · EMNIST seen classes (0-9) | log-likelihood: 1.01 \pm 0.00 -> 1.54 \pm 0.01 | Helped |
| Use TNP-A rather than the strongest NP baseline | Image completion · EMNIST unseen classes (10-46) | log-likelihood: 0.94 \pm 0.01 -> 1.41 \pm 0.01 | Helped |
| Use TNP-D rather than the strongest NP baseline | Contextual bandit · Wheel bandit, average across \delta | cumulative regret: 19.17 \pm 21.36 -> 3.87 \pm 2.91 | Helped |
| Replace autoregressive decoding with a diagonal covariance (ablation) | TNP-A · 1-D regression, RBF | log-likelihood: 1.63 \pm 0.00 -> 1.39 \pm 0.00 | Hurt |
| Replace autoregressive decoding with a non-diagonal Gaussian covariance (ablation) | TNP-A · 1-D regression, RBF | log-likelihood: 1.63 \pm 0.00 -> 1.46 \pm 0.00 | Hurt |
| Embed context pairs and target inputs separately (ablation) | TNP-D · 1-D regression, RBF | log-likelihood: 1.388 \pm 0.004 -> 1.375 \pm 0.004 | Hurt |
| Replace coupled context-pair input with cross-attention (ablation) | TNP-D · 1-D regression, RBF | log-likelihood: 1.388 \pm 0.004 -> 0.115 \pm 0.337 | Hurt |
| Encode context and targets separately, without context-target interaction (ablation) | TNP-A · 1-D regression, RBF | log-likelihood: 1.628 \pm 0.001 -> 1.477 \pm 0.016 | Hurt |
Closest papers by meaning
- End-to-End Meta-Bayesian Optimisation with Transformer Neural Processes (NeurIPS 2023 · similarity 0.89)
- Test Time Scaling for Neural Processes (NeurIPS 2025 · similarity 0.88)
- Dimension Agnostic Neural Processes (ICLR 2025 · similarity 0.87)
- Translation Equivariant Transformer Neural Processes (ICML 2024 · similarity 0.86)
- Gridded Transformer Neural Processes for Spatio-Temporal Data (ICML 2025 · similarity 0.86)
- Rényi Neural Processes (ICML 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.