Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
BuildingsBench: A Large-Scale Dataset of 900K Buildings and Benchmark for Short-Term Load Forecasting
Patrick Emami, Abhijeet Sahu, Peter Graf
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 |
|---|---|---|---|
| pretrain on Buildings-900K (zero-shot) versus persistence ensemble | Transformer-L (Gaussian) · real commercial buildings | NRMSE: 16.68 -> 13.31 | Improved |
| pretrain on Buildings-900K (zero-shot) versus persistence ensemble | Transformer-L (Gaussian) · real residential buildings | NRMSE: 77.88 -> 79.34 | Hurt |
| fine-tune pretrained model on 6 months of target building data (ablation) | Transformer-L (Gaussian) · real commercial buildings | NRMSE: 13.03 -> 12.96 | Improved |
| fine-tune pretrained model on target building data (ablation) | Transformer-L (Gaussian) · real residential buildings | NRMSE: 79.43 -> 77.20 | Improved |
| fine-tune pretrained model on target building data (ablation) | Transformer-L (Tokens) · real residential buildings | NRMSE: 94.11 -> 94.53 | Hurt |
| pretrain on Buildings-900K before fine-tuning instead of training from scratch (ablation) | Transformer-L (Gaussian) · real commercial buildings | NRMSE: 37.21 -> 12.96 | Improved |
| pretrained + fine-tuned model versus LightGBM | Transformer-L (Gaussian) · real commercial buildings | NRMSE: 16.02 -> 12.96 | Improved |
| use Gaussian outputs instead of load tokens (ablation) | Transformer-L · real commercial buildings zero-shot | NRMSE: 14.46 -> 13.31 | Improved |
| scale pretraining data from 1K to 100K/900K buildings (ablation) | Transformer-M · real commercial buildings | NRMSE | Improved |
| scale pretraining data (ablation) | Transformer-M · real residential buildings | NRMSE | No clear effect |
| increase model size from M to L (ablation) | Transformer · real commercial buildings | NRMSE | No clear effect |
| fine-tune only the last layer instead of all layers (ablation) | pretrained transformers · real buildings transfer | NRMSE | Hurt |
Closest papers by meaning
- NeuralBES: A Differentiable, Control-Aware Emulator for Scalable Building Energy Modeling (NeurIPS 2026 · similarity 0.87)
- Building Timeseries Dataset: Empowering Large-Scale Building Analytics (NeurIPS 2024 · similarity 0.87)
- PF∆: A Benchmark Dataset for Power Flow under Load, Generation, and Topology Variations (NeurIPS 2025 · similarity 0.83)
- Revisiting the Generic Transformer: Deconstructing a Strong Baseline for Time Series Foundation Models (NeurIPS 2026 · similarity 0.83)
- ProbTS: Benchmarking Point and Distributional Forecasting across Diverse Prediction Horizons (NeurIPS 2024 · similarity 0.82)
- Sundial: A Family of Highly Capable Time Series Foundation Models (ICML 2025 · similarity 0.82)
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.
