Pointwise Metrics Mislead: An Evaluation Protocol for Multimodal Inverse Problems

Mads H. Baattrup, Jörn Bach, Laurids Jeppe, Finn Labe, Alexander Grohsjean, Christian Schwanenberger, Peer Stelldinger

arXiv:2605.22891 · 2026-05-25 공개 · arXiv · PDF

model-ranking uncertainty-calibration particle-physics crps-evaluation spectrum-fidelity scientific-reconstruction posterior-distribution inverse-problem-evaluation

Abstract

Evaluation in scientific reconstruction is dominated by pointwise metrics - RMSE, MAE, per-event resolution - under the implicit assumption that lower error means better reconstruction. We show that this assumption fails structurally for inverse problems with multimodal posteriors. By the law of total variance, point estimators trained to minimize MSE or MAE produce a marginal spectrum strictly narrower than the truth whenever the posterior has nonzero width. The resulting bias is independent of architecture, training, and dataset size, and it compresses precisely the spectral features - tails, modes, shapes - that downstream scientific measurements rely on. We propose a three-part evaluation protocol where each step targets a failure mode the others miss: per-event distributional accuracy via CRPS, population-level marginal accuracy via a spectrum-fidelity diagnostic, and uncertainty trustworthiness via coverage-based calibration. On a synthetic benchmark with an analytic posterior and on a realistic many-to-one inverse problem from particle physics, model rankings reverse between pointwise and distributional metrics, and calibration further separates architectures indistinguishable under CRPS. The evaluation protocol, not the model, determines the scientific conclusion.

한국어 요약

📋 한 줄 요약

**[과학 재구성 평가 / 다봉 사후분포]** Pointwise metric(RMSE·MAE)이 multimodal posterior 역문제에서 구조적으로 실패함을 입증, CRPS·spectrum fidelity·calibration 3단 평가 프로토콜 제시로 model ranking이 reversal됨을 보임.

🎯 핵심 기여도

💡 핵심 아이디어

Multimodal posterior 역문제에서는 pointwise metric이 모델 자체가 아닌 evaluation protocol에 의해 과학적 결론이 결정되며, 분포 정확도·spectral fidelity·calibration을 동시에 검사해야 신뢰성 있는 평가가 가능하다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 과학 재구성 분야의 평가 표준에 근본적 재고를 요구, multimodal posterior 역문제 평가의 원리적 프로토콜 제공, downstream 과학 측정의 신뢰성 회복 경로 제시. **한계**: 입자물리 사례 중심으로 다른 과학 도메인(천체·기후·의료) 일반화 추가 검증 필요, CRPS·spectrum diagnostic 계산 비용, calibration 검정의 sample size 의존.

🚀 실용적 활용