Machine Learning Sensors (ML Sensors)

This repository contains resources for the ML sensor datasheet template introduced in Datasheets for Machine Learning Sensors. An ML sensor is an embedded edge device that encapsulates sensing, on-device ML processing, and data governance at the hardware boundary, while exposing a thin, sensor-like interface to the rest of the system. Learn more at: mlsensors.org.

Repository metadata (FAIR and reuse)

This repository is designed to support Findable, Accessible, Interoperable, and Reusable (FAIR) usage of the datasheet template and example instantiations.

• License: See LICENSE
• How to cite: See CITATION.cff (or CITATION)
• Authorship and contact: See AUTHORS.md and the contact email below
• Versioning and change history: See CHANGELOG.md and the version history section in the template

Contact:

Quick start (use the template)

1. Open the editable template: ml_sensor_datasheet_template.md
2. Duplicate it and fill it in for your ML sensor.
3. Export to PDF if desired (optional): pandoc or your preferred Markdown-to-PDF tool.
4. Submit improvements via pull request. Please include a brief note describing the change and update the version history.

The ML sensor datasheet

The ML sensor datasheet provides a structured overview of the hardware, data, model, privacy/security, and system-level behavior relevant to an ML sensor. It builds on and unifies prior documentation approaches that describe parts of the pipeline, including:

• Datasheets for datasets
• Data nutrition labels
• Model cards
• Traditional hardware sensor datasheets

The template covers topics including (not limited to): environmental impact, communication protocols, device diagrams, dataset attributes, model characteristics, end-to-end performance analysis, compliance and certifications, maintenance/versioning, and hardware specifications. The goal is to improve transparency, explainability, and auditability of ML sensors in real deployment contexts.

Template and example datasheets (reproducibility)

To support reproducibility and adoption, this repository includes:

• An editable, empty template (Markdown):
  ml_sensor_datasheet_template.md
• Two completed example datasheets using the full template:
  • Open-source ML sensor datasheet (PDF):
    open_source_datasheet_updated.pdf
  • Commercial ML sensor datasheet (PDF):
    commercial_datasheet_updated.pdf

Versioning and updates

ML sensors may evolve over time due to firmware updates, model retraining, calibration changes, or hardware revisions. We therefore treat datasheets as versioned artifacts:

• Each release should specify the hardware/firmware/model version it documents.
• Updates should increment the datasheet version and summarize changes.
• When end-to-end performance changes, updated measurements should be reported and tied to the updated version.

Contributing

Contributions are welcome. Please open an issue or submit a pull request.

• If you modify the template, update the template version history.
• If you add a new datasheet instance, include the filled template and (optionally) a PDF export.
• If you add figures or photos, include caption provenance (for example, “Photo by authors” or a source link).