Fairness (machine Learning) on:  
[Wikipedia]  
[Google]  
[Amazon]
Fairness in
machine learning refers to the various attempts at correcting
algorithmic bias
Algorithmic bias describes systematic and repeatable errors in a computer system that create " unfair" outcomes, such as "privileging" one category over another in ways different from the intended function of the algorithm.
Bias can emerge from ...
in automated decision processes based on machine learning models. Decisions made by computers after a machine-learning process may be considered unfair if they were based on
variables considered sensitive. Examples of these kinds of variable include
gender,
ethnicity,
sexual orientation,
disability and more. As it is the case with many ethical concepts, definitions of fairness and bias are always controversial. In general, fairness and bias are considered relevant when the decision process impacts people's lives. In
machine learning, the problem of
algorithmic bias
Algorithmic bias describes systematic and repeatable errors in a computer system that create " unfair" outcomes, such as "privileging" one category over another in ways different from the intended function of the algorithm.
Bias can emerge from ...
is well known and well studied. Outcomes may be skewed by a range of factors and thus might be considered unfair with respect to certain groups or individuals. An example would be the way social media sites deliver personalized news to consumers.
Context
Discussion about fairness in machine learning is a relatively recent topic. Since 2016 there has been a sharp increase in research into the topic. This increase could be partly accounted to an influential report by
ProPublica
ProPublica (), legally Pro Publica, Inc., is a nonprofit organization based in New York City. In 2010, it became the first online news source to win a Pulitzer Prize, for a piece written by one of its journalists''The Guardian'', April 13, 2010P ...
that claimed that the
COMPAS
Compas, also known as compas direct or compas direk (; Haitian Creole: ''konpa'', ''kompa'' or ''kompa dirèk''), is a modern méringue dance music genre of Haiti. The genre was popularized following the creation of Ensemble Aux Callebasses in ...
software, widely used in US courts to predict
recidivism, was racially biased.
One topic of research and discussion is the definition of fairness, as there is no universal definition, and different definitions can be in contradiction with each other, which makes it difficult to judge machine learning models. Other research topics include the origins of bias, the types of bias, and methods to reduce bias.
In recent years tech companies have made tools and manuals on how to detect and reduce
bias
Bias is a disproportionate weight ''in favor of'' or ''against'' an idea or thing, usually in a way that is closed-minded, prejudicial, or unfair. Biases can be innate or learned. People may develop biases for or against an individual, a group, ...
in machine learning.
IBM has tools for
Python
Python may refer to:
Snakes
* Pythonidae, a family of nonvenomous snakes found in Africa, Asia, and Australia
** ''Python'' (genus), a genus of Pythonidae found in Africa and Asia
* Python (mythology), a mythical serpent
Computing
* Python (pr ...
and
R with several algorithms to reduce software bias and increase its fairness.
Google has published guidlines and tools to study and combat bias in machine learning.
Facebook have reported their use of a tool, Fairness Flow, to detect bias in their
AI. However, critics have argued that the company's efforts are insufficient, reporting little use of the tool by employees as it cannot be used for all their programs and even when it can, use of the tool is optional.
Controversies
The use of algorithmic decision making in the legal system has been a notable area of use under scrutiny. In 2014, then
U.S. Attorney General
The United States attorney general (AG) is the head of the United States Department of Justice, and is the chief law enforcement officer of the federal government of the United States. The attorney general serves as the principal advisor to the p ...
Eric Holder raised concerns that "risk assessment" methods may be putting undue focus on factors not under a defendant's control, such as their education level or socio-economic background. The 2016 report by
ProPublica
ProPublica (), legally Pro Publica, Inc., is a nonprofit organization based in New York City. In 2010, it became the first online news source to win a Pulitzer Prize, for a piece written by one of its journalists''The Guardian'', April 13, 2010P ...
on
COMPAS
Compas, also known as compas direct or compas direk (; Haitian Creole: ''konpa'', ''kompa'' or ''kompa dirèk''), is a modern méringue dance music genre of Haiti. The genre was popularized following the creation of Ensemble Aux Callebasses in ...
claimed that black defendants were almost twice as likely to be incorrectly labelled as higher risk than white defendants, while making the opposite mistake with white defendants.
The creator of
COMPAS
Compas, also known as compas direct or compas direk (; Haitian Creole: ''konpa'', ''kompa'' or ''kompa dirèk''), is a modern méringue dance music genre of Haiti. The genre was popularized following the creation of Ensemble Aux Callebasses in ...
, Northepointe Inc., disputed the report, claiming their tool is fair and ProPublica made statistical errors, which was subsequently refuted again by ProPublica.
Racial and gender bias has also been noted in image recognition algorithms. Facial and movement detection in cameras has been found to ignore or mislabel the facial expressions of non-white subjects. In 2015, the automatic tagging feature in both
Flickr and
Google Photos was found to label black people with tags such as "animal" and "gorilla". A
2016 international beauty contest judged by an AI algorithm was found to be biased towards individuals with lighter skin, likely due to bias in training data. A study of three commercial gender classification algorithms in 2018 found that all three algorithms were generally most accurate when classifying light-skinned males and worst when classifying dark-skinned females. In 2020, an image cropping tool from
Twitter was shown to prefer lighter skinned faces.
DALL-E, a machine learning
Text-to-image model released in 2021, has been prone to create racist and sexist images that reinforce societal stereotypes, something that has been admitted by its creators.
Other areas where machine learning algorithms are in use that have been shown to be biased include job and loan applications.
Amazon
Amazon most often refers to:
* Amazons, a tribe of female warriors in Greek mythology
* Amazon rainforest, a rainforest covering most of the Amazon basin
* Amazon River, in South America
* Amazon (company), an American multinational technology co ...
has used software to review job applications that was sexist, for example by penalizing resumes that included the word "women". In 2019,
Apple's algorithm to determine credit card limits for their new
Apple Card
Apple Card is a credit card created by Apple Inc. and issued by Goldman Sachs, designed primarily to be used with Apple Pay on Apple devices such as an iPhone, iPad, Apple Watch, or Mac. Currently, it is available only in the United States, wit ...
gave significantly higher limits to males than females, even for couples that shared their finances. Mortgage-approval algorithms in use in the U.S. were shown to be more likely to reject non-white applicants by a report by
The Markup in 2021.
Group Fairness criteria
In
classification Classification is a process related to categorization, the process in which ideas and objects are recognized, differentiated and understood.
Classification is the grouping of related facts into classes.
It may also refer to:
Business, organizat ...
problems, an algorithm learns a function to predict a discrete characteristic
, the target variable, from known characteristics
. We model
as a discrete
random variable which encodes some characteristics contained or implicitly encoded in
that we consider as sensitive characteristics (gender, ethnicity, sexual orientation, etc.). We finally denote by
the prediction of the
classifier.
Now let us define three main criteria to evaluate if a given classifier is fair, that is if its predictions are not influenced by some of these sensitive variables.
[Solon Barocas; Moritz Hardt; Arvind Narayanan]
''Fairness and Machine Learning''
Retrieved 15 December 2019.
Independence
We say the
random variables
satisfy independence if the sensitive characteristics
are
statistically independent of the prediction
, and we write
We can also express this notion with the following formula:
This means that the classification rate for each target classes is equal for people belonging to different groups with respect to sensitive characteristics
.
Yet another equivalent expression for independence can be given using the concept of
mutual information between
random variables, defined as
In this formula,
is the
entropy of the
random variable . Then
satisfy independence if
.
A possible
relaxation of the independence definition include introducing a positive
slack
Finally, another possible
relaxation is to require
.
Separation
We say the
random variables
satisfy separation if the sensitive characteristics
are
statistically independent of the prediction
given the target value
, and we write
We can also express this notion with the following formula:
This means that all the dependence of the decision
on the sensitive attribute
must be justified by the actual dependence of the true target variable
.
Another equivalent expression, in the case of a binary target rate, is that the
true positive rate
''Sensitivity'' and ''specificity'' mathematically describe the accuracy of a test which reports the presence or absence of a condition. Individuals for which the condition is satisfied are considered "positive" and those for which it is not are ...
and the
false positive rate are equal (and therefore the
false negative rate and the
true negative rate are equal) for every value of the sensitive characteristics:
A possible relaxation of the given definitions is to allow the value for the difference between rates to be a
positive number lower than a given
slack is a measure of the distance (at a given level of the probability score) between the ''predicted'' cumulative percent negative and ''predicted'' cumulative percent positive.
The greater this separation coefficient is at a given score value, the more effective the model is at differentiating between the set of positives and negatives at a particular probability cut-off. According to Mayes: "It is often observed in the credit industry that the selection of validation measures depends on the modeling approach. For example, if modeling procedure is parametric or semi-parametric, the
two-sample K-S test is often used. If the model is derived by heuristic or iterative search methods, the measure of model performance is usually
divergence. A third option is the coefficient of separation...The coefficient of separation, compared to the other two methods, seems to be most reasonable as a measure for model performance because it reflects the separation pattern of a model."
Sufficiency
We say the
random variables
satisfy sufficiency if the sensitive characteristics
are
statistically independent of the target value
given the prediction
, and we write
We can also express this notion with the following formula:
This means that the
probability of actually being in each of the groups is equal for two individuals with different sensitive characteristics given that they were predicted to belong to the same group.
Relationships between definitions
Finally, we sum up some of the main results that relate the three definitions given above:
* Assuming
is binary, if
and
are not
statistically independent, and
and
are not
statistically independent either, then independence and separation cannot both hold.
* If
as a
joint distribution has positive
probability for all its possible values and
and
are not
statistically independent, then separation and sufficiency cannot both hold.
Mathematical formulation of group fairness definitions
Preliminary definitions
Most statistical measures of fairness rely on different metrics, so we will start by defining them. When working with a
binary
Binary may refer to:
Science and technology Mathematics
* Binary number, a representation of numbers using only two digits (0 and 1)
* Binary function, a function that takes two arguments
* Binary operation, a mathematical operation that ta ...
classifier, both the predicted and the actual classes can take two values: positive and negative. Now let us start explaining the different possible relations between predicted and actual outcome:
[Verma, Sahil, and Julia Rubin]
"Fairness definitions explained."
In 2018 IEEE/ACM international workshop on software fairness (fairware), pp. 1-7. IEEE, 2018.
* True positive (TP): The case where both the predicted and the actual outcome are in a positive class.
* True negative (TN): The case where both the predicted outcome and the actual outcome are assigned to the negative class.
* False positive (FP): A case predicted to befall into a positive class assigned in the actual outcome is to the negative one.
* False negative (FN): A case predicted to be in the negative class with an actual outcome is in the positive one.
These relations can be easily represented with a
confusion matrix
In the field of machine learning and specifically the problem of statistical classification, a confusion matrix, also known as an error matrix, is a specific table layout that allows visualization of the performance of an algorithm, typically a su ...
, a table that describes the accuracy of a classification model. In this matrix, columns and rows represent instances of the predicted and the actual cases, respectively.
By using these relations, we can define multiple metrics which can be later used to measure the fairness of an algorithm:
* Positive predicted value (PPV): the fraction of positive cases which were correctly predicted out of all the positive predictions. It is usually referred to as
precision, and represents the
probability of a correct positive prediction. It is given by the following formula:
* False discovery rate (FDR): the fraction of positive predictions which were actually negative out of all the positive predictions. It represents the
probability of an erroneous positive prediction, and it is given by the following formula:
* Negative predicted value (NPV): the fraction of negative cases which were correctly predicted out of all the negative predictions. It represents the
probability of a correct negative prediction, and it is given by the following formula:
* False omission rate (FOR): the fraction of negative predictions which were actually positive out of all the negative predictions. It represents the
probability of an erroneous negative prediction, and it is given by the following formula:
* True positive rate (TPR): the fraction of positive cases which were correctly predicted out of all the positive cases. It is usually referred to as sensitivity or recall, and it represents the
probability of the positive subjects to be classified correctly as such. It is given by the formula:
* False negative rate (FNR): the fraction of positive cases which were incorrectly predicted to be negative out of all the positive cases. It represents the
probability of the positive subjects to be classified incorrectly as negative ones, and it is given by the formula:
* True negative rate (TNR): the fraction of negative cases which were correctly predicted out of all the negative cases. It represents the
probability of the negative subjects to be classified correctly as such, and it is given by the formula:
* False positive rate (FPR): the fraction of negative cases which were incorrectly predicted to be positive out of all the negative cases. It represents the
probability of the negative subjects to be classified incorrectly as positive ones, and it is given by the formula:
The following criteria can be understood as measures of the three general definitions given at the beginning of this section, namely Independence, Separation and Sufficiency. In the table
to the right, we can see the relationships between them.
To define these measures specifically, we will divide them into three big groups as done in Verma et al.:
definitions based on a predicted outcome, on predicted and actual outcomes, and definitions based on predicted probabilities and the actual outcome.
We will be working with a binary classifier and the following notation:
refers to the score given by the classifier, which is the probability of a certain subject to be in the positive or the negative class.
represents the final classification predicted by the algorithm, and its value is usually derived from
, for example will be positive when
is above a certain threshold.
represents the actual outcome, that is, the real classification of the individual and, finally,
denotes the sensitive attributes of the subjects.
Definitions based on predicted outcome
The definitions in this section focus on a predicted outcome
for various
distributions of subjects. They are the simplest and most intuitive notions of fairness.
* Demographic parity, also referred to as statistical parity, acceptance rate parity and benchmarking. A classifier satisfies this definition if the subjects in the protected and unprotected groups have equal probability of being assigned to the positive predicted class. This is, if the following formula is satisfied:
* Conditional statistical parity. Basically consists in the definition above, but restricted only to a
subset of the instances. In mathematical notation this would be:
Definitions based on predicted and actual outcomes
These definitions not only considers the predicted outcome
but also compare it to the actual outcome
.
* Predictive parity, also referred to as outcome test. A classifier satisfies this definition if the subjects in the protected and unprotected groups have equal PPV. This is, if the following formula is satisfied:
: Mathematically, if a classifier has equal PPV for both groups, it will also have equal FDR, satisfying the formula:
* False positive error rate balance, also referred to as predictive equality. A classifier satisfies this definition if the subjects in the protected and unprotected groups have aqual FPR. This is, if the following formula is satisfied:
: Mathematically, if a classifier has equal FPR for both groups, it will also have equal TNR, satisfying the formula:
* False negative error rate balance, also referred to as equal opportunity. A classifier satisfies this definition if the subjects in the protected and unprotected groups have equal FNR. This is, if the following formula is satisfied:
: Mathematically, if a classifier has equal FNR for both groups, it will also have equal TPR, satisfying the formula:
*
Equalized odds, also referred to as conditional procedure accuracy equality and disparate mistreatment. A classifier satisfies this definition if the subjects in the protected and unprotected groups have equal TPR and equal FPR, satisfying the formula:
* Conditional use accuracy equality. A classifier satisfies this definition if the subjects in the protected and unprotected groups have equal PPV and equal NPV, satisfying the formula:
* Overall accuracy equality. A classifier satisfies this definition if the subject in the protected and unprotected groups have equal prediction accuracy, that is, the probability of a subject from one class to be assigned to it. This is, if it satisfies the following formula:
* Treatment equality. A classifier satisfies this definition if the subjects in the protected and unprotected groups have an equal ratio of FN and FP, satisfying the formula:
Definitions based on predicted probabilities and actual outcome
These definitions are based in the actual outcome
and the predicted probability score
.
* Test-fairness, also known as calibration or matching conditional frequencies. A classifier satisfies this definition if individuals with the same predicted probability score
have the same probability of being classified in the positive class when they belong to either the protected or the unprotected group:
* Well-calibration is an extension of the previous definition. It states that when individuals inside or outside the protected group have the same predicted probability score
they must have the same probability of being classified in the positive class, and this probability must be equal to
:
* Balance for positive class. A classifier satisfies this definition if the subjects constituting the positive class from both protected and unprotected groups have equal average predicted probability score
. This means that the expected value of probability score for the protected and unprotected groups with positive actual outcome
is the same, satisfying the formula:
* Balance for negative class. A classifier satisfies this definition if the subjects constituting the negative class from both protected and unprotected groups have equal average predicted probability score
. This means that the expected value of probability score for the protected and unprotected groups with negative actual outcome
is the same, satisfying the formula:
Social welfare function
Some scholars have proposed defining algorithmic fairness in terms of a
social welfare function
In welfare economics, a social welfare function is a function that ranks social states (alternative complete descriptions of the society) as less desirable, more desirable, or indifferent for every possible pair of social states. Inputs of the ...
. They argue that using a social welfare function enables an algorithm designer to consider fairness and predictive accuracy in terms of their benefits to the people affected by the algorithm. It also allows the designer to
trade off efficiency and equity in a principled way.
Sendhil Mullainathan has stated that algorithm designers should use social welfare functions in order to recognize absolute gains for disadvantaged groups. For example, a study found that using a decision-making algorithm in
pretrial detention
Remand, also known as pre-trial detention, preventive detention, or provisional detention, is the process of detaining a person until their trial after they have been arrested and charged with an offence. A person who is on remand is held i ...
rather than pure human judgment reduced the detention rates for Blacks, Hispanics, and racial minorities overall, even while keeping the crime rate constant.
Individual Fairness criteria
An important distinction among fairness definitions is the one between group and individual notions.
[Mehrabi, Ninareh, Fred Morstatter, Nripsuta Saxena, ]Kristina Lerman
Kristina Lerman (born 1967) is an American network scientist whose research concerns the spread of information on social networks, and fairness in machine learning. She is a research professor at the University of Southern California, in the Co ...
, and Aram Galstyan
"A survey on bias and fairness in machine learning."
ACM Computing Surveys (CSUR) 54, no. 6 (2021): 1-35. Roughly speaking, while group fairness criteria compare quantities at a group level, typically identified by sensitive attributes (e.g. gender, ethnicity, age, etc...), individual criteria compare individuals. In words, individual fairness follow the principle that "similar individuals should receive similar treatments".
There is a very intuitive approach to fairness, which usually goes under the name of Fairness Through Unawareness (FTU), or ''Blindness'', that prescribe not to explicitly employ sensitive features when making (automated) decisions. This is effectively a notion of individual fairness, since two individuals differing only for the values of their sensitive attributes would receive the same outcome.
However, in general, FTU is subject to several drawbacks, the main being that it does not take into account possible correlations between sensitive attributes and non-sensitive attributes employed in the decision-making process. For example, an agent with the (malignant) intention to discriminate on the basis of gender could introduce in the model a proxy variable for gender (i.e. a variable highly correlated with gender) and effectively using gender information while at the same time being compliant to the FTU prescription.
The problem of ''what variables correlated to sensitive ones are fairly employable by a model'' in the decision-making process is a crucial one, and is relevant for
group concepts as well: independence metrics require a complete removal of sensitive information, while separation-based metrics allow for correlation, but only as far as the labeled target variable "justify" them.
The most general concept of individual fairness was introduced in the pioneer work by Dwork and collaborators in 2012 and can be thought of as a mathematical translation of the principle that the decision map taking features as input should be built such that it is able to "map similar individuals similarly", that is expressed as a
Lipschitz condition
In mathematical analysis, Lipschitz continuity, named after German mathematician Rudolf Lipschitz, is a strong form of uniform continuity for functions. Intuitively, a Lipschitz continuous function is limited in how fast it can change: there exi ...
on the model map. They call this approach Fairness Through Awareness (FTA), precisely as counterpoint to FTU, since they underline the importance of choosing the appropriate target-related distance metric in order to assess which individuals are ''similar'' in specific situations. Again, this problem is very related to the point raised above about what variables can be seen as "legitimate" in particular contexts.
Causality-based metrics
Causal fairness measures the frequency with which two nearly identical users or applications who differ only in a set of characteristics with respect to which resource allocation must be fair receive identical treatment.
An entire branch of the academic research on fairness metrics is devoted to leverage causal models to assess bias in
machine learning models. This approach is usually justified by the fact that the same observational distribution of data may hide different causal relationships among the variables at play, possibly with different interpretations of whether the outcome are affected by some form of bias or not.
Kusner et al.
[Kusner, M. J., Loftus, J., Russell, C., & Silva, R. (2017)]
Counterfactual fairness
Advances in neural information processing systems, 30. propose to employ
counterfactuals
Counterfactual conditionals (also ''subjunctive'' or ''X-marked'') are conditional sentences which discuss what would have been true under different circumstances, e.g. "If Peter believed in ghosts, he would be afraid to be here." Counterfactual ...
, and define a decision-making process counterfactually fair if, for any individual, the outcome does not change in the counterfactual scenario where the sensitive attributes are changed. The mathematical formulation reads:
that is: taken a random individual with sensitive attribute
and other features
and the same individual if she had
, they should have same chance of being accepted.
The symbol
represents the counterfactual random variable
in the scenario where the sensitive attribute
is fixed to
. The conditioning on
means that this requirement is at the individual level, in that we are conditioning on all the variables identifying a single observation.
Machine learning models are often trained upon data where the outcome depended on the decision made at that time. For example, if a machine learning model has to determine whether an inmate will recidivate and will determine whether the inmate should be released early, the outcome could be dependent on whether the inmate was released early or not. Mishler et al.
propose a formula for counterfactual equalized odds:
where
is a random variable,
denotes the outcome given that the decision
was taken, and
is a sensitive feature.
Bias Mitigation strategies
Fairness can be applied to machine learning algorithms in three different ways:
data preprocessing Data preprocessing can refer to manipulation or dropping of data before it is used in order to ensure or enhance performance, and is an important step in the data mining process. The phrase "garbage in, garbage out" is particularly applicable to ...
,
optimization during software training, or post-processing results of the algorithm.
Preprocessing
Usually, the classifier is not the only problem; the
dataset A data set (or dataset) is a collection of data. In the case of tabular data, a data set corresponds to one or more database tables, where every column of a table represents a particular variable, and each row corresponds to a given record of th ...
is also biased. The discrimination of a dataset
with respect to the group
can be defined as follows:
That is, an approximation to the difference between the probabilities of belonging in the positive class given that the subject has a protected characteristic different from
and equal to
.
Algorithms correcting bias at preprocessing remove information about dataset variables which might result in unfair decisions, while trying to alter as little as possible. This is not as simple as just removing the sensitive variable, because other attributes can be correlated to the protected one.
A way to do this is to map each individual in the initial dataset to an intermediate representation in which it is impossible to identify whether it belongs to a particular protected group while maintaining as much information as possible. Then, the new representation of the data is adjusted to get the maximum accuracy in the algorithm.
This way, individuals are mapped into a new multivariable representation where the probability of any member of a protected group to be mapped to a certain value in the new representation is the same as the probability of an individual which doesn't belong to the protected group. Then, this representation is used to obtain the prediction for the individual, instead of the initial data. As the intermediate representation is constructed giving the same probability to individuals inside or outside the protected group, this attribute is hidden to the classificator.
An example is explained in Zemel et al.
[Richard Zemel; Yu (Ledell) Wu; Kevin Swersky; Toniann Pitassi; Cyntia Dwork]
''Learning Fair Representations''
Retrieved 1 December 2019 where a
multinomial random variable is used as an intermediate representation. In the process, the system is encouraged to preserve all information except that which can lead to biased decisions, and to obtain a prediction as accurate as possible.
On the one hand, this procedure has the advantage that the preprocessed data can be used for any machine learning task. Furthermore, the classifier does not need to be modified, as the correction is applied to the
dataset A data set (or dataset) is a collection of data. In the case of tabular data, a data set corresponds to one or more database tables, where every column of a table represents a particular variable, and each row corresponds to a given record of th ...
before processing. On the other hand, the other methods obtain better results in accuracy and fairness.
[Ziyuan Zhong]
''Tutorial on Fairness in Machine Learning''
Retrieved 1 December 2019
Reweighing
Reweighing is an example of a preprocessing algorithm. The idea is to assign a weight to each dataset point such that the weighted
discrimination
Discrimination is the act of making unjustified distinctions between people based on the groups, classes, or other categories to which they belong or are perceived to belong. People may be discriminated on the basis of Racial discrimination, r ...
is 0 with respect to the designated group.
[Faisal Kamiran; Toon Calders]
''Data preprocessing techniques for classification without discrimination''
Retrieved 17 December 2019
If the dataset
was unbiased the sensitive variable
and the target variable
would be
statistically independent and the probability of the
joint distribution would be the product of the probabilities as follows:
In reality, however, the dataset is not unbiased and the variables are not
statistically independent so the observed probability is:
To compensate for the bias, the software adds a
weight, lower for favored objects and higher for unfavored objects. For each
we get:
When we have for each
a weight associated
we compute the weighted discrimination with respect to group
as follows:
It can be shown that after reweighting this weighted discrimination is 0.
Inprocessing
Another approach is to correct the
bias
Bias is a disproportionate weight ''in favor of'' or ''against'' an idea or thing, usually in a way that is closed-minded, prejudicial, or unfair. Biases can be innate or learned. People may develop biases for or against an individual, a group, ...
at training time. This can be done by adding constraints to the optimization objective of the algorithm.
[Muhammad Bilal Zafar; Isabel Valera; Manuel Gómez Rodríguez; Krishna P. Gummadi]
''Fairness Beyond Disparate Treatment & Disparate Impact: Learning Classification without Disparate Mistreatment''
Retrieved 1 December 2019 These constraints force the algorithm to improve fairness, by keeping the same rates of certain measures for the protected group and the rest of individuals. For example, we can add to the objective of the
algorithm
In mathematics and computer science, an algorithm () is a finite sequence of rigorous instructions, typically used to solve a class of specific problems or to perform a computation. Algorithms are used as specifications for performing ...
the condition that the false positive rate is the same for individuals in the protected group and the ones outside the protected group.
The main measures used in this approach are false positive rate, false negative rate, and overall misclassification rate. It is possible to add just one or several of these constraints to the objective of the algorithm. Note that the equality of false negative rates implies the equality of true positive rates so this implies the equality of opportunity. After adding the restrictions to the problem it may turn intractable, so a relaxation on them may be needed.
This technique obtains good results in improving fairness while keeping high accuracy and lets the
programmer
A computer programmer, sometimes referred to as a software developer, a software engineer, a programmer or a coder, is a person who creates computer programs — often for larger computer software.
A programmer is someone who writes/creates ...
choose the fairness measures to improve. However, each machine learning task may need a different method to be applied and the code in the classifier needs to be modified, which is not always possible.
Adversarial debiasing
We train two
classifiers at the same time through some gradient-based method (f.e.:
gradient descent). The first one, the ''predictor'' tries to accomplish the task of predicting
, the target variable, given
, the input, by modifying its weights
to minimize some
loss function . The second one, the ''adversary'' tries to accomplish the task of predicting
, the sensitive variable, given
by modifying its weights
to minimize some loss function
.
[Brian Hu Zhang; Blake Lemoine; Margaret Mitchell]
''Mitigating Unwanted Biases with Adversarial Learning''
Retrieved 17 December 2019[Joyce Xu]
''Algorithmic Solutions to Algorithmic Bias: A Technical Guide''
Retrieved 17 December 2019
An important point here is that, in order to propagate correctly,
above must refer to the raw output of the classifier, not the discrete prediction; for example, with an
artificial neural network
Artificial neural networks (ANNs), usually simply called neural networks (NNs) or neural nets, are computing systems inspired by the biological neural networks that constitute animal brains.
An ANN is based on a collection of connected units ...
and a classification problem,
could refer to the output of the
softmax layer.
Then we update
to minimize
at each training step according to the
gradient and we modify
according to the expression:
where
is a tuneable
hyperparameter
In Bayesian statistics, a hyperparameter is a parameter of a prior distribution; the term is used to distinguish them from parameters of the model for the underlying system under analysis.
For example, if one is using a beta distribution to mo ...
that can vary at each time step.
The intuitive idea is that we want the ''predictor'' to try to minimize
(therefore the term
) while, at the same time, maximize
(therefore the term
), so that the ''adversary'' fails at predicting the sensitive variable from
.
The term
prevents the ''predictor'' from moving in a direction that helps the ''adversary'' decrease its loss function.
It can be shown that training a ''predictor'' classification model with this algorithm improves
demographic parity with respect to training it without the ''adversary''.
Postprocessing
The final method tries to correct the results of a classifier to achieve fairness. In this method, we have a classifier that returns a score for each individual and we need to do a binary prediction for them. High scores are likely to get a positive outcome, while low scores are likely to get a negative one, but we can adjust the
threshold to determine when to answer yes as desired. Note that variations in the threshold value affect the trade-off between the rates for true positives and true negatives.
If the score function is fair in the sense that it is independent of the protected attribute, then any choice of the threshold will also be fair, but classifiers of this type tend to be biased, so a different threshold may be required for each protected group to achieve fairness.
A way to do this is plotting the true positive rate against the false negative rate at various threshold settings (this is called ROC curve) and find a threshold where the rates for the protected group and other individuals are equal.
[Moritz Hardt; Eric Price; Nathan Srebro]
''Equality of Opportunity in Supervised Learning''
Retrieved 1 December 2019
The advantages of postprocessing include that the technique can be applied after any classifiers, without modifying it, and has a good performance in fairness measures. The cons are the need to access to the protected attribute in test time and the lack of choice in the balance between accuracy and fairness.
Reject Option based Classification
Given a
classifier let
be the probability computed by the classifiers as the
probability that the instance
belongs to the positive class +. When
is close to 1 or to 0, the instance
is specified with high degree of certainty to belong to class + or - respectively. However, when
is closer to 0.5 the classification is more unclear.
[Faisal Kamiran; Asim Karim; Xiangliang Zhang]
''Decision Theory for Discrimination-aware Classification''
Retrieved 17 December 2019
We say
is a "rejected instance" if
with a certain
such that
.
The algorithm of "ROC" consists on classifying the non-rejected instances following the rule above and the rejected instances as follows: if the instance is an example of a deprived group (
) then label it as positive, otherwise, label it as negative.
We can optimize different measures of
discrimination
Discrimination is the act of making unjustified distinctions between people based on the groups, classes, or other categories to which they belong or are perceived to belong. People may be discriminated on the basis of Racial discrimination, r ...
(link) as functions of
to find the optimal
for each problem and avoid becoming discriminatory against the privileged group.
See also
*
Algorithmic bias
Algorithmic bias describes systematic and repeatable errors in a computer system that create " unfair" outcomes, such as "privileging" one category over another in ways different from the intended function of the algorithm.
Bias can emerge from ...
*
Machine learning
References
Machine learning
Information ethics
Computing and society
Philosophy of artificial intelligence
Discrimination
Bias