Algorithmic Transparency via Quantitative Input Influence

By A. Datta. 2017

55 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Management Science and Operations Research, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Management Science and Operations Research

In Safety, Robustness & Interpretability

Citations in the corpus

Towards the Science of Security and Privacy in Machine Learning (2016, background): “measures to estimate the influence of specific inputs on the model output [112].”
European Union Regulations on Algorithmic Decision-Making and a "Right to Explanation" (2016, background): “One promising avenue of research concerns developing algorithms to quantify the degree of influence of input variables on outputs, given black-box access to a trained prediction algorithm [11].”
SoK: Security and Privacy in Machine Learning (2018, background): “introduced quantitative input influence measures to estimate the influence of specific inputs on the model output [112].”
The ethics of algorithms: Mapping the debate (2016, neutral): “Algorithms can only be considered explainable to the degree that a human can articulate the trained model or rationale of a particular decision, for instance by explaining the (quantiﬁed) inﬂuence of particular inputs or attributes (Datta et al., 2016).”
Privacy Risks of Explaining Machine Learning Models (2019, methodology): “In the category of perturbation-based methods fall occlusion based methods [42], LIME [29] which trains a local model, and QII [12] which computes the Shapley value of each feature.”
Preventing undesirable behavior of intelligent machines (2019, background): “For example, research has proposed means for determining how sensitive a black-box supervised learning algorithm is to each feature in a vector of features used to represent the input [105, 106], which can be used to determine the impact of features, such as race and gender [107].”
EU regulations on algorithmic decision-making and a "right to explanation" (2016, background): “Research is underway in pursuit of rendering algorithms more amenable to ex post and ex ante inspection (Datta et al. , 2016; Vellido et al., 2012). Furthermore, a number of recent studies have attempted to tackle the issue of discrimination within algorithms by introducing tools t o both identify (”
FairTest: Discovering Unwarranted Associations in Data-Driven Applications (2015, background): “Understanding and identifying unwarranted associations is an important step towards holding automated decision-making entities accountable for unfair practices, thus also providing incentive for the adoption of corrective measures [1], [2], [6], [7].”
Alignment for Advanced Machine Learning Systems (2020, background): “In a related vein, Datta, Sen, and Zick (2016), ˇStrumbelj and Kononenko (2014), and Robnik-ˇSikonja and Kononenko (2008) deﬁne metrics for reporting the inﬂuence of various inputs and sets of inputs on the output of a black-box ML system.”
The Price of Interpretability (2019, background): “Others deﬁne metrics of feature importance to quantify the effect of each feature in the overall model [17, 11].”
Auditing Search Engines for Differential Satisfaction Across Demographics (2017, methodology): “Other work sought to detect unfairness through reverse A/B testing with synthetic users [31, 12], algorithmic auditing [37], and analysis of predictions made by supervised machine learning methods [24].”
On Ensuring that Intelligent Machines Are Well-Behaved (2017, background): “…proposed means for determining how sensitive a black-box supervised learning algorithm is to each feature in a vector of features used to represent the input [Datta et al., 2016; Adler et al., 2016], which can be used to determine the impact of features like race and gender [Datta et al., 2015].”
Fairness-Aware Programming (2019, background): “• The past few years have witnessed numerous techniques for constructing fair classifiers [1, 4, 11, 13, 25], as well as for testing and auditing fairness of existing decision-making procedures [2, 8, 12, 23].”