Catalogue of Tools & Metrics for Trustworthy AI

These tools and metrics are designed to help AI actors develop and use trustworthy AI systems and applications that respect human rights and are fair, transparent, explainable, robust, secure and safe.

Scope

SUBMIT A METRIC

If you have a tool that you think should be featured in the Catalogue of AI Tools & Metrics, we would love to hear from you!

SUBMIT
This page includes technical metrics and methodologies for measuring and evaluating AI trustworthiness and AI risks. These metrics are often represented through mathematical formulas that assess the technical requirements for achieving trustworthy AI in a particular context. They can help to ensure that a system is fair, accurate, explainable, transparent, robust, safe, or secure.

Hellinger distance

The Hellinger distance (sometimes called the Jeffreys distance) is a metric in the space of probability distributions. The Hellinger distance can be used to quantify the degree of similarity between two probability ...

Objectives:


The demographic disparity metric (DD) determines whether a facet has a larger proportion of the rejected outcomes in the dataset than of the accepted outcomes. In the binary case where there are two facets, men and women for example, that constitute the dat...

Objectives:


RADio introduces a rank-aware Jensen Shannon (JS) divergence. This combination accounts for (i) a user’s decreasing propensity to observe items further down a list and (ii) full distributional shifts as opposed to point estimates.

Objectives:


Contextual Outlier INterpretation (COIN) is a method designed to explain the abnormality of existing outliers spotted by detectors. The interpretability for an outlier is achieved from three aspects: outlierness score, att that contribute to the abnormality, a...

Objectives:


Given an input data sample, LEMNA generates a small set of interpretable features to explain how the input sample is classified. The core idea is to approximate a local area of the complex deep learning decision boundary using a simple interpretable model. The...

Objectives:


SHAP (SHapley Additive exPlanations) assigns each feature an importance value for a particular prediction. Its novel components include: (1) the identification of a new class of additive feature importance measures, and (2) theoretical results showing there is...

LIME is a novel explanation technique that explains the predictions of any classifier in an interpretable and faithful manner, by learning an interpretable model locally around the prediction.

Objectives:


Following the VIC framework, our proposed ShapleyVIC extends the widely used Shapley-based variable importance measures beyond final models for a comprehensive assessment and has important practical implications.

Objectives:


Ideally we would like to obtain a more complete understanding of variable importance for the set of models that predict almost equally well. This set of almost-equally-accurate predictive models is called the Rashomon set; it is the set of models with training...

CLIPBERTSCORE is a simple weighted combination of CLIPScore (Hessel et al., 2021) and BERTScore (Zhang* et al., 2020) to leverage the robustness and strong factuality detection performance between image-summary and document-summary, respectively.

Objectives:


The Banzhaf power index is a power index defined by the probability of changing an outcome of a vote where voting rights are not necessarily equally divided among the voters. Data Banzhaf uses this notion to measure data points' "voting powers" towards algorit...

Objectives:


Beta Shapley is a unified data valuation framework that naturally arises from Data Shapley by relaxing the efficiency axiom. The Beta(α, β)-Shapley value considers the pair of hyperparameters (α, β) which decides the weight distribution on [n]. Beta(1,1)-Shapl...

In a cooperative game, there are n players D = {1,...,n} and a score function v : 2[n] → R assigns a reward to each of 2 n subsets of players: v(S) is the reward if the players in subset S ⊆ D cooperate. We view the supervised machine learning problem as a coo...

Objectives:


The Surrogacy Efficacy Score is a technique for gaining a better understanding of the inner workings of complex "black box" models. For example, by using a Tree-based model, this method provides a more interpretable representation of the model’s behavior by...


The Partial Dependence Complexity metric uses the concept of Partial Dependence curve to evaluate how simple this curve can be represented. The partial dependence curve is used to show model predictions are affected on average by each feature. Curves repres...


The α-Feature Importance metric quantifies the minimum proportion of features required to represent α of the total importance. In other words, this metric is focused in obtaining the minimum number of features necessary to obtain no less than α × 100% of th...


The PGC metric compares the top-K ranking of features importance drawn from the entire dataset with the top-K ranking induced from specific subgroups of predictions. It can be applied to both categorical and regression problems, being useful for quantifying...


Local Feature Importance refers to the assignment of feature normalized importance to different regions of the input data space. For a given dataset D with N samples, it is possible to compute a vector of feature importance for each individual observation d...


The metric GFIS is based on the concept of entropy. More precisely on the entropy of the normalized features measure, which represents the concentration of information within a set of features. Lower entropy values indicate that the majority of the explanat...

Objectives:


Machine learning models, at the core of AI applications,  typically achieve a high accuracy at the expense of an insufficient explainability. Moreover, according to the proposed regulations,  AI applications based on machine learning must be "trus...


catalogue Logos
Sign up for OECD artificial intelligence newsletter