arxiv2026-06-29arXiv:2606.30919

Budget-Adaptive Routing: Skipping the Weak When the Strong Answers Anyway

Wei Geng, Nitinder Mohan, Jörg Ott

実装難易度

Hard

推論・学習コスト

Medium

想定用途

検出

PaperGitHubgithub.dev で開く →

概要

Abstract

Edge-cloud inference collaborations are often designed with a routing estimator that decides whether to offload each frame from weak models at the edge to stronger models in the cloud. Existing systems place the routing estimator after the weak detector, so the weak forward pass still runs even on frames that are later offloaded. In this paper, we argue that this weak-conditioned design can be

何が新しいか

Edge-cloud inference collaborations are often designed with a routing estimator that decides whether to offload each frame from weak models at the edge to stronger models in the cloud. Existing…

何に使えるか

検出

実装情報

Paper URL
あり
GitHub URL
あり

実装チェックリスト

実装または配布ページ

OK

コードまたはモデル配布ページから検証を始められます。

一次情報リンク

OK

Paper / GitHub

検証しやすさ

要確認

大規模モデル、生成、強化学習など再現コストが高い要素を含みます。

計算資源

未取得

推論中心なら軽めですが、再学習時はGPUが必要になる可能性があります。

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。

製造業適性 12
  1. 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. 2LightGBMやRandom Forestなどのベースラインを先に作り、この手法と比較します。
  3. 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。
  4. 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 - 大規模モデル、生成、強化学習など再現コストが高い要素を含みます。

必要リソース

  • GPU目安: Medium
  • 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。
  • 推論中心なら軽めですが、再学習時はGPUが必要になる可能性があります。

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