Course contentsShow
← Machine Learning and Deep Learning
Lesson 3475 of 3,53876. Ethics and Societal ImpactFree lesson

Reporting and Transparency in ML Emissions

Learn best practices for documenting and communicating the environmental impact of ML projects.

Reporting and Transparency in ML Emissions

What you'll learn: How to document and communicate your machine learning project's environmental footprint using standardized practices and transparent reporting frameworks.

Why Transparency Matters

Once you've measured your model's energy consumption and carbon footprint (from previous lessons), the next ethical obligation is sharing that information. Transparency in ML emissions serves multiple purposes: it enables peer comparison, drives accountability, informs stakeholders, and accelerates community-wide progress toward sustainable AI.

Think of it like nutrition labels on food—standardized reporting lets everyone make informed decisions and compare options fairly.

Core Reporting Practices

Carbon Emissions Statements: Include a dedicated section in papers, model cards, or documentation that reports:

  • Total energy consumed during training (kWh)
  • Carbon emissions (CO2 equivalent)
  • Hardware used (GPU/TPU types and quantities)
  • Training duration and location (regional grid carbon intensity varies)
  • Number of training runs and experiments

Model Cards Extension: Extend traditional model cards to include environmental metrics alongside performance metrics. This contextualizes the environmental cost against model capabilities.

Standardized Frameworks: Use tools like the ML CO2 Impact calculator or CodeCarbon that generate consistent, comparable reports. These provide structured outputs you can share directly.

Comparative Context: Don't just report absolute numbers—provide context. Compare to baseline models, previous versions, or equivalent tasks. State whether you used carbon offsets or renewable energy.

Transparency in Practice

Be honest about uncertainties in your measurements. Report both training and expected inference costs when possible. If you optimized for efficiency, describe those choices. If environmental impact wasn't prioritized, acknowledge that gap.

Key Takeaway: Transparent, standardized reporting of ML emissions—including energy consumption, carbon footprint, hardware, and context—is essential for accountability and enables the research community to collectively pursue more sustainable AI practices.