Toward Calibrated Mixture-of-Experts Under Distribution Shift

When a model says it is 80% confident, it should be right about 80% of the time — that is calibration, and it matters enormously in high-stakes settings like medicine, finance, and autonomous systems. Mixture-of-experts architectures, which route inputs to specialized sub-models, have shown strong performance gains, but their calibration behavior under real-world distribution shift has been poorly understood. This paper fills that gap, revealing when and why expert-level calibration fails to propagate to the full model. The proposed adversarial reweighting fix has practical implications for deploying robust, trustworthy MoE systems in production environments where training and deployment data inevitably diverge.