Description
This issue proposes upgrading the core prediction engine to a Multi-Model AI Ensemble Engine. Currently, Stream A and Stream B outputs are fused via a fixed linear combination.
We will implement dynamic weighting based on input quality metrics: if the eye crop is dark or blurry, its ensemble weight is automatically scaled down and the body/gill weights are scaled up. We will also implement confidence intervals using logit distribution spread, and add fallback logic.
Technical Implementation Details
- Dynamic Weights: Modify
main.py to calculate dynamic stream weights.
- Confidence Interval: Compute margin of error from probability deviations.
- Graceful Degradation: Recalibrate weights if a stream fails to load.
Description
This issue proposes upgrading the core prediction engine to a Multi-Model AI Ensemble Engine. Currently, Stream A and Stream B outputs are fused via a fixed linear combination.
We will implement dynamic weighting based on input quality metrics: if the eye crop is dark or blurry, its ensemble weight is automatically scaled down and the body/gill weights are scaled up. We will also implement confidence intervals using logit distribution spread, and add fallback logic.
Technical Implementation Details
main.pyto calculate dynamic stream weights.