Which approach is a strategy for mitigating bias in CPMAI?

• A. Model ensembling and hyperparameter tuning.
• B. Data diversification and reweighting, algorithmic fairness techniques, and ongoing auditing.
• C. Data sweeping and cursor auditing.
• D. Relying solely on a single data source.

Answer: (B)

Bias mitigation in CPMAI involves addressing both the data feeding the model and how the model behaves over time. Diversifying data sources and reweighting the data helps counteract underrepresentation and prevents the model from learning biased patterns that come from skewed input. Applying algorithmic fairness techniques provides concrete methods to constrain or adjust the model’s decisions so they treat different groups more equitably. Ongoing auditing keeps an eye on performance and fairness after deployment, catching drift and triggering corrective actions as the system operates in the real world. Together, these elements form a comprehensive approach to reduce bias throughout the project lifecycle.

Relying on a single data source tends to amplify bias, while focusing only on tuning models for accuracy via ensembling or hyperparameters doesn't inherently address fairness concerns. Data sweeping or cursory auditing without the fairness tools and ongoing checks is insufficient for sustained bias mitigation.