As algorithms increasingly inform and influence decisions made about individuals, it
becomes increasingly important to address concerns that these algorithms might be
discriminatory. The output of an algorithm can be discriminatory for many reasons, most
notably: (1) the data used to train the algorithm might be biased (in various ways) to favor
certain populations over others; (2) the analysis of this training data might inadvertently or
maliciously introduce biases that are not borne out in the data. This work focuses on the
latter concern.

We develop and study multicalbration: a new measure of algorithmic fairness that aims to
mitigate concerns about discrimination that is introduced in the process of learning a
predictor from data. Multicalibration guarantees accurate (calibrated) predictions for every
subpopulation that can be identified within a specified class of computations. We think of
the class as being quite rich, in particular it can contain many and overlapping subgroups of a
protected group.

We show that in many settings this strong notion of protection from discrimination is both
attainable and aligned with the goal of obtaining accurate predictions. Along the way, we
present new algorithms for learning a multicalibrated predictor, study the computational
complexity of this task, and draw new connections to computational learning models such as
agnostic learning.

Joint work with Ursula Hebert-Johnson, Michael P. Kim and Guy Rothblum. Online version: https://arxiv.org/pdf/1711.08513.pdf