All Tags AWS ai algorithm-design architecture browser cloud cloud-efficiency cloud-principles cost-reduction data-centric data-compression data-processing deployment design documentation edge-computing email-sharing energy-efficiency energy-footprint enterprise-optimization green-ai hardware libraries llm locality machine-learning maintainability management measured microservices migration mobile model-optimization model-training multi-objective network-traffic parameter-tuning performance queries rebuilding scaling services storage-optimization strategies tabs template testing workloads

Tactic: Choose an energy efficient drift detection algorithm

Tactic sort: Awesome Tactic

Type: Software Practice

Category: green-ml-enabled-systems

Tags: energy-footprint machine-learning measured model-training

Title

Choose an energy efficient drift detection algorithm

Description

Select concept drift detection algorithms with energy efficiency in mind, ensuring a balanced trade off between detection accuracy and energy consumption. Since drift detectors themselves consume energy and may trigger retraining procedures, false positives from low accuracy detectors can lead to unnecessary retraining, thereby increasing the overall energy footprint. Therefore, when concept drift checks are infrequent or retraining is significantly more energy consuming than detection, it may be preferable to use more accurate detectors even if they are energy intensive (e.g., KSWIN for abrupt drift). Conversely, in high frequency check settings, detectors with moderate energy use and sufficient accuracy, such as HDDM_W or ADWIN, can reduce redundant computation and maintain overall sustainability.

Participant

ML researchers and practitioners

Related software artifact

Concept drift detectors

Context

ML enabled systems

Software feature

Dynamic model updating

Tactic intent

To reduce the overall energy consumption of adaptive machine learning systems by energy-aware choices when selecting concept drift detection algorithms

Target quality attribute

Energy efficiency

Other related quality attributes

Accuracy

Measured impact

The maximum energy consumption disparity observed between detectors reached 53.2%, with ADWIN consuming considerably less energy than KSWIN for similar accuracy levels

Source

Rafiullah Omar; Justus Bogner; Joran Leest; Vincenzo Stoico; Patricia Lago; Henry Muccini (2024) How to Sustainably Monitor ML-Enabled Systems? Accuracy and Energy Efficiency Tradeoffs in Concept Drift Detection (DOI: https://ieeexplore.ieee.org/abstract/document/10805468)