The k-anonymity privacy requirement for publishing microdata requires that each equivalence class (i.e., a set of records are indistinguishable from other with respect to certain "identifying" attributes) contains at least k records. Recently, several authors have recognized cannot prevent attribute disclosure. notion l-diversity has been proposed address this; l well-represented values sensitive attribute. In this paper we show number limitations. particular, it is neither necessary nor...

What does it mean for an algorithm to be biased? In U.S. law, unintentional bias is encoded via disparate impact, which occurs when a selection process has widely different outcomes groups, even as appears neutral. This legal determination hinges on definition of protected class (ethnicity, gender) and explicit description the process.

A key goal of the fair-ML community is to develop machine-learning based systems that, once introduced into a social context, can achieve and legal outcomes such as fairness, justice, due process. Bedrock concepts in computer science---such abstraction modular design---are used define notions fairness discrimination, produce fairness-aware learning algorithms, intervene at different stages decision-making pipeline "fair" outcomes. In this paper, however, we contend that these render...

Computers are increasingly used to make decisions that have significant impact on people's lives. Often, these predictions can affect different population subgroups disproportionately. As a result, the issue of fairness has received much recent interest, and number fairness-enhanced classifiers appeared in literature. This paper seeks study following questions: how do techniques fundamentally compare one another, what accounts for differences? Specifically, we seek bring attention many...

What does it mean for an algorithm to be fair? Different papers use different notions of algorithmic fairness, and although these appear internally consistent, they also seem mutually incompatible. We present a mathematical setting in which the distinctions previous can made formal. In addition characterizing spaces inputs (the "observed" space) outputs "decision" space), we introduce notion construct space: space that captures unobservable, but meaningful variables prediction. show order...

What does it mean to be fair?

The k-anonymity privacy requirement for publishing microdata requires that each equivalence class (i.e., a set of records are indistinguishable from other with respect to certain "identifying" attributes) contains at least k records. Recently, several authors have recognized cannot prevent attribute disclosure. notion ℓ-diversity has been proposed address this; ℓ well-represented (in Section 2) values sensitive attribute. In this paper, we show number limitations. particular, it is neither...

RF sensor networks are wireless that can localize and track people (or targets) without needing them to carry or wear any electronic device. They use the change in received signal strength (RSS) of links due movements infer their locations. In this paper, we consider real-time multiple target tracking with networks. We apply radio tomographic imaging (RTI), which generates images propagation field, as if they were frames a video. Our RTI method uses RSS measurements on frequency channels...

Network radio frequency (RF) environment sensing (NRES) systems pinpoint and track people in buildings using changes the signal strength measurements made by a wireless sensor network. It has been shown that such can locate who do not participate system wearing any device, even through walls, because of moving cause to static However, many cannot stationary people. We present evaluate which or people, without calibration, kernel distance quantify difference between two histograms...

Predictive policing systems are increasingly used to determine how allocate police across a city in order best prevent crime. Discovered crime data (e.g., arrest counts) help update the model, and process is repeated. Such have been empirically shown be susceptible runaway feedback loops, where repeatedly sent back same neighborhoods regardless of true rate. In response, we develop mathematical model predictive that proves why this loop occurs, show exhibits such problems, demonstrate change...

Philosophers have established that certain ethically important values are modally robust in the sense they systematically deliver correlative benefits across a range of counterfactual scenarios. In this paper, we contend recourse - systematic process reversing unfavorable decisions by algorithms and bureaucracies scenarios is such good. particular, argue two essential components good life temporally extended agency trust underwritten recourse.

Game-theoretic formulations of feature importance have become popular as a way to "explain" machine learning models. These methods define cooperative game between the features model and distribute influence among these input elements using some form game's unique Shapley values. Justification for rests on two pillars: their desirable mathematical properties, applicability specific motivations explanations. We show that problems arise when values are used solutions mitigate necessarily induce...

In most algorithmic applications which compare two distributions, information theoretic distances are more natural than standard lp norms. this paper we design streaming and sublinear time property testing algorithms for entropy various distances.Batu et al posed the problem of with respect to Jensen-Shannon distance. We present optimal estimating bounded, symmetric f-divergences (including divergence Hellinger distance) between distributions in frameworks. Along way, close a (log n)/H gap...

The geometric median is a classic robust estimator of centrality for data in Euclidean spaces. In this paper we formulate the on Riemannian manifold as minimizer sum geodesic distances to points. We prove existence and uniqueness manifolds with non-positive sectional curvature give sufficient conditions positively curved manifolds. Generalizing Weiszfeld procedure finding data, present an algorithm computing arbitrary manifold. show that converges unique solution when it exists. This method...