In a recent webinar I had asked about potential solutions to support balancing across multiple objectives and speaker ran out of time before answering. It occurs to me that there is a lot of interesting prior work from classical learning domain associated with balancing between multiple (potentially divergent) objectives as aggregated into a single meta metric, particularly in context of classification involving balancing between bias / variance tradeoffs, which extend simple accuracy metrics to include alternate framings like AUC, f1 score, etc. I am wondering if such forms of aggregated metrics could be packaged into a simple Ocean packet for integration into a packaged form of objective?
In a recent webinar I had asked about potential solutions to support balancing across multiple objectives and speaker ran out of time before answering. It occurs to me that there is a lot of interesting prior work from classical learning domain associated with balancing between multiple (potentially divergent) objectives as aggregated into a single meta metric, particularly in context of classification involving balancing between bias / variance tradeoffs, which extend simple accuracy metrics to include alternate framings like AUC, f1 score, etc. I am wondering if such forms of aggregated metrics could be packaged into a simple Ocean packet for integration into a packaged form of objective?