Uncertainty Quantification
Uncertainty Quantification · UQ
Reporting, as a number, how much a prediction can be trusted.
In materials terms
Ranking candidates by predicted band gap with an error range attached.
Materials & AI · glossary
Materials and AI glossary
Met an unfamiliar term while reading? Start with the concept.
10 terms
See allThe examples explain the concept. For actual research results, see AI in materials에서 원문과 함께 확인하세요. 엄선한 용어 86개에는 Colab에서 바로 돌려볼 수 있는 실습 노트북이 있습니다. “AI 자동 추가” 표시가 붙은 용어는 새 이슈를 수집할 때 자동으로 추가된 것으로, 아직 사람이 검토하지 않았고 실습 노트북이 없습니다. 노트북의 데이터는 개념 확인용으로 생성한 값이며 실제 측정값이 아닙니다.
scikit-learn 가우시안 프로세스 문서 ↗
Aleatoric and Epistemic Uncertainty
Aleatoric · Epistemic Uncertainty
The distinction between uncertainty that remains however much data you add, such as measurement noise, and uncertainty that comes from having too little data and shrinks as you gather more.
In materials terms
Separating the scatter that repeat measurements do not reduce from the uncertainty of an unexplored region.
Prediction Interval
Prediction Interval · 신뢰 구간
The range a new observation is expected to fall in. It is wider than a confidence interval on the mean.
In materials terms
Reporting a predicted strength together with a 90 percent interval.
Calibration
Calibration · 신뢰도 곡선 · Reliability Diagram
Checking whether the probabilities or intervals a model reports match the observed frequencies, and correcting them if not.
In materials terms
Confirming on validation data that 90 percent of observations really fall inside the 90 percent interval.
Sensitivity Analysis
Sensitivity Analysis · OAT · Morris
Quantifying how much each input variable affects the output.
Sobol Index
Sobol Index · Sobol Indices · 분산 기반 전역 민감도
A global sensitivity measure obtained by decomposing the output variance into the contributions of the inputs and their combinations. The first-order index is a variable acting alone; the total index includes its interactions.
In materials terms
Keeping only the two of five process variables with a large total index for detailed experiments.
Monte Carlo Sampling
Monte Carlo Sampling · MC · 준몬테카를로
Estimating the distribution and the statistics of an output by drawing the inputs at random many times.
In materials terms
Feeding in the input error distributions to obtain the distribution of the predicted property.
Error Propagation
Error Propagation · 불확실성 전파
Estimating how uncertainty in the inputs carries through a calculation into uncertainty in the output.
In materials terms
Seeing how far an error in measured thickness grows in the calculated conductivity.
Feature Importance
Feature Importance · Permutation Importance · 순열 중요도
A value indicating which inputs matter most to a model’s performance or its predictions. It does not imply causation.
In materials terms
Finding the descriptor whose shuffling raises the error the most.
SHAP Values
SHAP · Shapley Value · 섀플리 값
Values that explain a prediction by dividing it into the contributions of the individual inputs, used to interpret a single prediction.
In materials terms
Seeing why the predicted strength came out high for one composition, contribution by contribution.