Skip to content

Towards Sustainability of AI

Main NavigationHome Design Patterns

Appearance

Sidebar Navigation

Sustainable development of ML

Introduction

The paper

Sustainability dimensions

🌳 Ecological

ML Demand Specification

Assess Performance-Efficiency Tradeoff

Decide on Environmental Infrastructure

Evaluate ML-Model-Hardware-Fit

Data Collection and Preparation

Promote Data Sparseness in Data Collection

Reduce Data Dimensionality

Compress Data Storage

Stage Preprocessed Data

Modeling and Training

Preselect Energy-Efficient ML Models

Eliminate Inefficiency in ML-Model Architecture

Streamline ML-Model Training Process

Optimize Hyperparameter Efficiently

Deployment and Monitoring

Streamline ML Retraining Frequency

Design Computationally Sparse ML Architecture

Report and Monitor Environmental Sustainability

😇 Social

ML Demand Specification

Assess Social Implications

Conceptualize Definition of Fairness

Define Human Role in Decision Process

Data Collection and Preparation

Foster Accurate and Fair Data Collection

Understand and Establish Fairness in Dataset

Leverage Fair Data Sampling

Modeling and Training

Leverage Fair and Interpretable Models

Conduct Fairness Evaluation

Adjust Model Parameters for Fairness

Deployment and Monitoring

Ensure Continuous (Human) Monitoring for Fairness

Convey ML-Model Understanding to End Users

Enhance Transparency through Fairness Metrics

🏛️ Governance

ML Demand Specification

Comply with Legal Frameworks and Company Policies

Compose Diverse and Interdisciplinary ML-Team

Establish a Responsible ML-Mindset

Promote ML Democratization

Data Collection and Preparation

Establish Standards in Data Collection and Preparation

Initiate Intra- and Interorganizational Data Democratization

Modeling and Training

Introduce ML-Model Transparency for Active Participation

Deployment and Monitoring

Ensure Documentation and Publishing

Engage in ML-Model Auditing

On this page

Table of Contents for current page

Towards Sustainability of AI – Identifying Design Patterns for Sustainable Machine Learning Development

Authors: Daniel Leuthe, Tim Meyer-Hollatz, Tobias Plank, Anja Senkmüller

Link: tbd.

Date of publishing: tbd.

Previous pageIntroduction

Next pageAssess Performance-Efficiency Tradeoff