Obsidian Source: Drafts / A Project

Summary

Pending synthesis from local Obsidian source.

Original source title: A Project

Extracted Preview

Evaluating biases in AI Systems

Conduct experiments for mitigating biases(data, algorithmic, demographic). We can use techniques like debiasing, adversarial training or diversification.

Drilling down into datasets(training biases - review data sampling for under-represented groups plus the way training data is labelled), algorithms(some factors might unintentionally discriminate) and other factors(cognitive biases - favoring some datasets over others).
Report : https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.1270.pdf
Python Package and Example Notebooks : https://github.com/Trusted-AI/AIF360
Potential : https://github.com/Trusted-AI/AIF360/blob/master/examples/demo_adversarial_debiasing.ipynb

Testing various ethical frameworks

Examine and compare various ethical frameworks for AI development - and assess their strengths, weakness and applicability. Testing models for prejudice, bias, social and environmental impact, data privacy, transparency, explainability - plus need affirmative evidence of that as well. Ethical Impact Assessment. There are ways to test systems on AI ethics - Scenario based testing, Data Analysis, Evaluation Metrics, Regulatory Compliance, Read Team Testing.

UNESCO recommended areas to focus - https://unesdoc.unesco.org/ark:/48223/pf0000385082.page=12
Adobe - https://www.adobe.com/content/dam/cc/en/ai-ethics/pdfs/Adobe-AI-Ethics-Principles.pdf (Responsibility, Accountability, Transparency)
Frameworks, Tools and Standards - https://github.com/EthicalML/awesome-artificial-intelligence-guidelines

I know what I will do in this one.

Explainable Reinforcement Learning

Integration Notes

Source folder: /home/yashs/Documents/Docs/Obsidian/Research-Notes
Local source: /home/yashs/Documents/Docs/Obsidian/Research-Notes/Drafts/A Project.md
Raw copy: raw/obsidian/research-notes/Drafts/A Project.md