[論文レビュー] Towards energy-efficient Deep Learning: An overview of energy-efficient approaches along the Deep Learning Lifecycle

Vanessa Mehlin, Sigurd Schacht+1 more|arXiv (Cornell University)|Feb 5, 2023

Green IT and Sustainability被引用数 18

ひとこと要約

この論文は、全体の Deep Learning ライフサイクルを通じたエネルギー効率のアプローチを調査し、IT infrastructure から deployment および evaluation まで、どこでエネルギー削減が可能かを説明します。

ABSTRACT

Deep Learning has enabled many advances in machine learning applications in the last few years. However, since current Deep Learning algorithms require much energy for computations, there are growing concerns about the associated environmental costs. Energy-efficient Deep Learning has received much attention from researchers and has already made much progress in the last couple of years. This paper aims to gather information about these advances from the literature and show how and at which points along the lifecycle of Deep Learning (IT-Infrastructure, Data, Modeling, Training, Deployment, Evaluation) it is possible to reduce energy consumption.

研究の動機と目的

• Deep Learning の環境影響の調査を促し、DLライフサイクルに沿ってエネルギーを節約できる場所を特定する。
• IT infrastructure, data, modeling, training, deployment, and evaluation にわたるエネルギー効率の高いアプローチの包括的概観を提供する。
• 既存の手法の分類学を提示し、文献の進展とギャップを強調する。

提案手法

• 最近の DL 研究に引用されたエネルギー効率のアプローチの文献調査を行う。
• DLライフサイクルの段階（IT infrastructure、data、modeling、training、deployment、evaluation）ごとにアプローチを整理・分類する。
• エネルギーを削減できる場所と発生し得るトレードオフを示すように、知見を統合する。

実験結果

リサーチクエスチョン

• RQ1 What energy-efficient approaches exist at each stage of the Deep Learning lifecycle?
• RQ2 How much energy can be saved by applying these approaches, and what are their trade-offs across stages (infrastructure, data, modeling, training, deployment, evaluation)?
• RQ3 What gaps remain in the literature and where is further research needed?

主な発見

• The literature shows progress in energy-efficient Deep Learning across multiple lifecycle stages.
• A taxonomy of energy-saving approaches is presented spanning IT infrastructure, data, modeling, training, deployment, and evaluation.
• The survey aggregates advances from recent years to map where reductions are achievable.