Sequential auditing of connected forecast errors under partial observations
Yuanbo Zhou, Junjie Xiong
実装難易度
Hard
推論・学習コスト
Low
想定用途
予測
概要
Graph forecasting systems often rely on prediction intervals to quantify uncertainty across connected locations, but the true outcomes needed for evaluation may be only partially observed. We study how an auditor can selectively retrieve missing outcomes to determine whether there exists a connected group whose outcomes violate the same side of their respective forecast intervals. The audit stops…
何が新しいか
Graph forecasting systems often rely on prediction intervals to quantify uncertainty across connected locations, but the true outcomes needed for evaluation may be only partially observed. We study…
何に使えるか
予測
実装情報
- Paper URL
- あり
実装チェックリスト
実装または配布ページ
要確認Paper onlyの可能性があるため再実装前提で確認してください。
一次情報リンク
OKPaper
検証しやすさ
要確認公式実装が見つからないため、論文から再実装する前提です。
計算資源
OK小規模データならCPUまたは単一GPUで検証しやすい領域です。
ライセンス
未取得配布元のLICENSE、モデルカード、Paperの利用条件を確認してください。
商用利用
未取得研究利用限定、データセット由来制限、API規約の有無を確認してください。
自社データで試すなら
製造業・材料開発のExcel/CSVデータに落とし込むための最初の手順です。
- 1まず自社データを、入力条件、目的変数、評価したい指標に分けて整理します。
- 2正常データだけで動くベースラインを作り、異常スコアのしきい値を現場知見と合わせます。
- 3評価指標はR2/RMSE、AUC、異常検知の再現率、実験回数削減率など、現場の意思決定に近いものを選びます。
- 4SHAPや特徴量重要度で、効いている因子が物理・化学・工程知識と矛盾しないか確認します。
実装難易度
Hard - 公式実装が見つからないため、論文から再実装する前提です。
必要リソース
- GPU目安: Low
- データセット: 論文・リポジトリ側の指定を確認してください。
- 学習要否: 再学習や評価環境の準備が必要になる可能性があります。
- 小規模データならCPUまたは単一GPUで検証しやすい領域です。
実務で使う場合の注意点
- ライセンスと商用利用条件は、Paper / GitHub / Hugging Face の配布元で確認してください。
- 精度、再現性、計算コストはデータセットや評価条件に依存します。
- 個人情報や機密データを扱う場合は、入力データの保存先と外部API利用条件を確認してください。
関連記事
TempShap: When and How Much a Feature Matters in Time-Series Forecasting
Attribution for a time-series forecaster should answer two questions that a feature-level scalar cannot: when
Unrolled gradients in disguise: bridging interpolation-based and Jacobian regularization for stable neural dynamics
Hybrid models and neural differential equations (NDEs) are becoming increasingly important for modelling physi
Hinted Scaling Laws
Scaling laws predict the outcome of large, expensive training runs by extrapolating trends from smaller, cheap
CoreNLP — CoreNLP: A Java suite of core NLP tools for tokenization, sentence segmentation, NER, parsing, coreference, sentiment analysis, etc.
CoreNLPはJavaで開発されたNLPツールのセットであり、分割、文分割、名詞認識、パーシング、コorefence、感情分析などを行える。