Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Publish best OrcaHello model(s) along with a RAIL (Responsible AI License) #150

Open
2 of 3 tasks
scottveirs opened this issue Jul 22, 2024 · 2 comments
Open
2 of 3 tasks
Labels
2024-hackathon Goals or issues for the 2024 annual Microsoft hackathon documentation Improvements or additions to documentation inference system Code to perform inference with the trained model(s) model training Data preparation and model training. question Further information is requested

Comments

@scottveirs
Copy link
Member

scottveirs commented Jul 22, 2024

This issue has three main tasks:

  • Publish the latest versions of the OrcaHello model (binary SRKW call classifier) within the aifororcas-livesystem repo, or -- if they exceed Github's 100MB/file limit -- in Orcasound's acoustic-sandbox S3 bucket
  • Document how each model was trained and deployed, and how it performs
  • Compose and publish with the model(s) a Responsible AI License (RAIL)

A recent RAIL example was published with Orcasound's orca-aye-eye` repo which presents an object detection algorithm for side-view images of boats and ships. The Orca Aye Eye RAIL was customized from boilerplate acquired via the Responsible AI license generator.

A key step in this final task will be for the OrcaHello model developers (past and current) and other interested Microsoft hackathon participants to compose an Attachment A with agreed-upon USE RESTRICTIONS... We can and should discuss potential restrictions below, but here are a few to ponder initially:

  1. Whaling industry
  2. Captive whale industry
  3. Whale-watching (commercial and recreational)
@scottveirs scottveirs added the 2024-hackathon Goals or issues for the 2024 annual Microsoft hackathon label Jul 22, 2024
@scottveirs
Copy link
Member Author

scottveirs commented Jul 22, 2024

Here is a first draft of a RAIL to cover models and associated apps & source code.

OrcaHello Real Time Inference System-RAIL.md

Where the Attachment A: USE RESTRICTIONS section at the end reads like this:

You agree to use the OrcaHello models to promote the conservation of scientific study of the endangered Southern Resident Killer Whales and other soniferous marine species. You agree not to use them in any way that further endangers them or other wildlife. Additionally, you agree not to use the Artifacts in furtherance of any of the following:

  1. Illegal or unethical whale watching behavior

    (a) In any way that violates any applicable national, federal, state, local or international law or regulation, including the U.S. Marine Mammal Protection Act and the rules issued by the Department of Fisheries and Oceans in Candaa.
    (b) In any way that results in vessels acting contrary to the Be Whale Wise voluntary guidance and regulations within the range of the endangered Southern Resident Killer Whales (from Northern BC to Northern California).
    (c) In any other ways that reduces the conservation of Southern Resident Killer Whales and other soniferous species, rather than increasing their conservation and recovery.

  2. Captive whale industry and whaling

    (a) In any way that supports the capture and captivity of soniferous whales, especially killer whales.
    (b) In connection with activities that present a risk of death or bodily harm to individual whales, including inciting or promoting violence, abuse, or any infliction of bodily harm to an individual or group of individuals.

Comments welcome!

@scottveirs scottveirs added documentation Improvements or additions to documentation question Further information is requested inference system Code to perform inference with the trained model(s) model training Data preparation and model training. labels Jul 22, 2024
@scottveirs
Copy link
Member Author

I've put a copy of the current model (.pkl file) along with the proposed RAIL and draft README.txt file inside a zip file named the same as the zip file containing the model in the Inference README.md. It has now been uploaded to the S3 bucket called acoustic sandbox that also contains other open models developed through Orcasound projects (Pod.Cast, OrcaAL, Orca-Eye-Aye, etc.)

By packaging the RAIL within the model's zip file, we can be more confident that our expectations for acceptable model use will be understood and observed.

Discuss the proposed RAIL and README here, and then vote on whether we should finalize them and take this approach.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
2024-hackathon Goals or issues for the 2024 annual Microsoft hackathon documentation Improvements or additions to documentation inference system Code to perform inference with the trained model(s) model training Data preparation and model training. question Further information is requested
Projects
Status: Todo
Development

No branches or pull requests

7 participants