Основные понятия
Conformal prediction sets' expected size is crucial for practical utility and can be theoretically quantified and empirically estimated.
Аннотация
The article discusses the importance of the expected size of conformal prediction sets and the challenges in quantifying it. It introduces theoretical quantification methods and empirical estimation procedures. The content is structured as follows:
- Abstract
- Introduction
- Background on Conformal Prediction Framework
- Theoretical Quantification of Expected Set Size
- Insights on Non-Conformity Function and User Parameters
- Conditional Expectation Analysis
- Practical Estimation Methods
- Experimental Results on Real-world Datasets
Статистика
"The expected size of the prediction sets under the split conformal prediction framework."
"The significance level α set to 0.1."
"The number of trials and success probability for the binomial random variable."
"The empirical approximation of ˜PR(r) for all r ∈ R."
Цитаты
"The sizes of the prediction sets are of significant relevance in practice."
"The split conformal framework satisfies Equation (1), but the quality of the prediction sets constructed depends on the data distribution and the machine learning model used."