Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
FIRE: Multi-Fidelity Regression with Distribution-Conditioned In-Context Learning Using Tabular Foundation Models
Rosen Ting-Ying Yu, Nicholas Sung, Faez Ahmed
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 |
|---|---|---|---|
| condition residual model on LF variance and quantiles, not only mean (ablation) | FIRE · 31 MF benchmarks | NRMSE Elo: about 250 Elo drop when removed | Improved |
| use TFM residual instead of scratch-trained GP (ablation) | FIRE (TabPFNv2.5 base) · 31 MF benchmarks | Elo: 774 -> 1000 | Improved |
| use TFM residual instead of random forest (ablation) | FIRE (TabPFNv2.5 base) · 31 MF benchmarks | Elo: 738 -> 1000 | Improved |
| use TFM residual instead of XGBoost (ablation) | FIRE (TabPFNv2.5 base) · 31 MF benchmarks | Elo: 605 -> 1000 | Improved |
| use one TFM per fidelity (recursive) instead of bi-level two surrogates (ablation) | FIRE · 31 MF benchmarks | Elo and runtime: over 10x runtime, no accuracy gain | No clear effect |
| use TFM instead of GP as base model (ablation) | FIRE · 31 MF benchmarks | NRMSE | Improved |
| replace TFM with GP in the FIRE framework (FIRE(GP)) (ablation) | FIRE · 31 MF benchmarks | Elo | Mixed |
| add post-hoc ensembling of the residual TFM (ablation) | FIRE · 31 MF benchmarks | NRMSE Elo: inference time increases by an order of magnitude | Improved |
| use Mitra instead of TabPFN as TFM backbone (ablation) | FIRE · 31 MF benchmarks | NRMSE | Hurt |
| use RealTabPFN instead of TabPFNv2.5 (ablation) | FIRE · 31 MF benchmarks | NRMSE | No clear effect |
| use FIRE on nested inputs rather than disjoint (ablation) | FIRE and GP baselines · synthetic benchmarks | Elo and rank | No clear effect |
| use single-pass TabPFNv2.5 with fidelity token instead of FT-PFN or LC-PFN | TabPFNv2.5 · LCBench | NRMSE | Improved |
Closest papers by meaning
- TabDPT: Scaling Tabular Foundation Models on Real Data (NeurIPS 2025 · similarity 0.84)
- Tabular Foundation Model for Generative Modelling (NeurIPS 2026 · similarity 0.84)
- Drift-Resilient TabPFN: In-Context Learning Temporal Distribution Shifts on Tabular Data (NeurIPS 2024 · similarity 0.84)
- Multi-Fidelity Residual Neural Processes for Scalable Surrogate Modeling (ICML 2024 · similarity 0.84)
- Uncertainty Quantification for Deep Regression using Contextualised Normalizing Flows (NeurIPS 2025 · similarity 0.83)
- A Unified Comparative Study with Generalized Conformity Scores for Multi-Output Conformal Regression (ICML 2025 · similarity 0.83)
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.
