Feb. 1 | Introduction, Course Logistics, Scheduling |
---|---|
Feb. 8 | |
Feb. 15 |
Yash presents 'Fides: Towards a Platform for Responsible Data Science'
|
Feb. 22 |
Shreya presents 'Algorithmic Decision Making and the Cost of Fairness'
|
March 1 |
Mridula presents 'Bias on the Web'
|
March 8 |
Xingtong presents 'Online Set Selection with Fairness and Diversity Constraints'
|
March 15 |
Wei presents 'A Nutritional Label for Rankings'
|
March 22 |
Bhavin presents 'Responsible Data Management'
|
March 29 | No class |
April 5 |
Xiaoxiao presents 'The Scored Society: Due Process for Automated Predictions'
|
April 12 | |
April 19 |
Mridula presents 'Ranking with Fairness Constraints'
|
April 26 | No One Signed Up (yet) |
May 3 | No One Signed Up (yet) |
Algorithmic decisions often result in scoring and ranking individuals to determine credit worthiness, qualifications for college admissions and employment, and compatibility as dating partners. While automatic and seemingly objective, ranking algorithms can discriminate against individuals and protected groups, and exhibit low diversity. Furthermore, ranked results are often unstable -- small changes in the input data or in the ranking methodology may lead to drastic changes in the output, making the result uninformative and easy to manipulate. Similar concerns apply in cases where items other than individuals are ranked, including colleges, academic departments, or products. Despite the ubiquity of rankers, there is, to the best of our knowledge, no technical work that focuses on making rankers transparent.
In this demonstration we present Ranking Facts, a Web-based application that generates a "nutritional label" for rankings. Ranking Facts is made up of a collection of visual widgets that implement our latest research results on fairness, stability, and transparency for rankings, and that communicate details of the ranking methodology, or of the output, to the end user. We will showcase Ranking Facts on real datasets from different domains, including college rankings, criminal risk assessment, and financial services.
We present a systematic approach for achieving fairness in a binary classification setting. While we focus on two well-known quantitative definitions of fairness, our approach encompasses many other previously studied definitions as special cases. The key idea is to reduce fair classification to a sequence of cost-sensitive classification problems, whose solutions yield a randomized classifier with the lowest (empirical) error subject to the desired constraints. We introduce two reductions that work for any representation of the cost-sensitive classifier and compare favorably to prior baselines on a variety of data sets, while overcoming several of their disadvantages.
Algorithms are now regularly used to decide whether defendants awaiting trial are too dangerous to be released back into the community. In some cases, black defendants are substantially more likely than white defendants to be incorrectly classified as high risk. To mitigate such disparities, several techniques have recently been proposed to achieve algorithmic fairness. Here we reformulate algorithmic fairness as constrained optimization: the objective is to maximize public safety while satisfying formal fairness constraints designed to reduce racial disparities. We show that for several past definitions of fairness, the optimal algorithms that result require detaining defendants above race-specific risk thresholds. We further show that the optimal unconstrained algorithm requires applying a single, uniform threshold to all defendants. The unconstrained algorithm thus maximizes public safety while also satisfying one important understanding of equality: that all individuals are held to the same standard, irrespective of race. Because the optimal constrained and unconstrained algorithms generally differ, there is tension between improving public safety and satisfying prevailing notions of algorithmic fairness. By examining data from Broward County, Florida, we show that this trade-off can be large in practice. We focus on algorithms for pretrial release decisions, but the principles we discuss apply to other domains, and also to human decision makers carrying out structured decision rules.
Data analysis impacts virtually every aspect of our society today. Often, this analysis is performed on an existing dataset, possibly collected through a process that the data scientists had limited control over. The existing data analyzed may not include the complete universe, but it is expected to cover the diversity of items in the universe. Lack of adequate coverage in the dataset can result in undesirable outcomes such as biased decisions and algorithmic racism, as well as creating vulnerabilities such as opening up room for adversarial attacks. In this paper, we assess the coverage of a given dataset over multiple categorical attributes. We first provide efficient techniques for traversing the combinatorial explosion of value combinations to identify any regions of attribute space not adequately covered by the data. Then, we determine the least amount of additional data that must be obtained to resolve this lack of adequate coverage. We confirm the value of our proposal through both theoretical analyses and comprehensive experiments on real data.
Data analysis impacts virtually every aspect of our society today. Often, this analysis is performed on an existing dataset, possibly collected through a process that the data scientists had limited control over. The existing data analyzed may not include the complete universe, but it is expected to cover the diversity of items in the universe. Lack of adequate coverage in the dataset can result in undesirable outcomes such as biased decisions and algorithmic racism, as well as creating vulnerabilities such as opening up room for adversarial attacks. In this paper, we assess the coverage of a given dataset over multiple categorical attributes. We first provide efficient techniques for traversing the combinatorial explosion of value combinations to identify any regions of attribute space not adequately covered by the data. Then, we determine the least amount of additional data that must be obtained to resolve this lack of adequate coverage. We confirm the value of our proposal through both theoretical analyses and comprehensive experiments on real data.
Data processing pipelines that are designed to clean, transform and alter data in preparation for learning predictive models, have an impact on those models’ accuracy and performance, as well on other properties, such as model fairness. It is therefore important to provide developers with the means to gain an in-depth understanding of how the pipeline steps affect the data, from the raw input to training sets ready to be used for learning. While other efforts track creation and changes of pipelines of relational operators, in this work we analyze the typical operations of data preparation within a machine learning process, and provide infrastructure for generating very granular provenance records from it, at the level of individual elements within a dataset. Our contributions include:(i) the formal definition of a core set of preprocessing operators,and the definition of provenance patterns for each of them, and(ii) a prototype implementation of an application-level provenance capture library that works alongside Python. We report on provenance processing and storage overhead and scalability experiments,carried out over both real ML benchmark pipelines and over TCP-DI, and show how the resulting provenance can be used to answer a suite of provenance benchmark queries that underpin some of the developers’ debugging questions, as expressed on the Data ScienceStack Exchange.
Issues of responsible data analysis and use are coming to the forefront of the discourse in data science research and practice, with most significant efforts to date on the part of the data mining, machine learning, and security and privacy communities. In these fields, the research has been focused on analyzing the fairness, accountability and transparency (FAT) properties of specific algorithms and their outputs. Although these issues are most apparent in the social sciences where fairness is interpreted in terms of the distribution of resources across protected groups, management of bias in source data affects a variety of fields. Consider climate change studies that require representative data from geographically diverse regions, or supply chain analyses that require data that represents the diversity of products and customers. Any domain that involves sparse or sampled data has exposure to potential bias.
In this vision paper, we argue that FAT properties must be considered as database system issues, further upstream in the data science lifecycle: bias in source data goes unnoticed, and bias may be introduced during preprocessing (fairness), spurious correlations lead to reproducibility problems (accountability), and assumptions made during preprocessing have invisible but significant effects on decisions (transparency). As machine learning methods continue to be applied broadly by non-experts, the potential for misuse increases. We see a need for a data sharing and collaborative analytics platform with features to encourage (and in some cases, enforce) best practices at all stages of the data science lifecycle. We describe features of such a platform, which we term Fides, in the context of urban analytics, outlining a systems research agenda in responsible data science.
Selection algorithms usually score individual items in isolation,and then select the top scoring items. However, often there is an additional diversity objective. Since diversity is a group property,it does not easily jibe with individual item scoring. In this paper,we study set selection queries subject to diversity and group fairness constraints. We develop algorithms for several problem settings with streaming data, where an online decision must be made on each item as it is presented. We show through experiments with real and synthetic data that fairness and diversity can be achieved, usually with modest costs in terms of quality.Our experimental evaluation leads to several important in-sights in online set selection. We demonstrate that theoretical guarantees on solution quality are conservative in real datasets,and that tuning the length of the score estimation phase leads to an interesting accuracy-efficiency trade-off. Further, we show that if a difference in scores is expected between groups, then these groups must be treated separately during processing. Otherwise, a solution may be derived that meets diversity constraints,but that selects lower-scoring members of disadvantaged groups
Ranking algorithms are deployed widely to order a set of items in applications such as search engines, news feeds, and recommendation systems. Recent studies, however, have shown that, left unchecked, the output of ranking algorithms can result in decreased diversity in the type of content presented, promote stereotypes, and polarize opinions. In order to address such issues, we study the following variant of the traditional ranking problem when, in addition, there are fairness or diversity constraints. Given a collection of items along with 1) the value of placing an item in a particular position in the ranking, 2) the collection of sensitive attributes (such as gender, race, political opinion) of each item and 3) a collection of fairness constraints that, for each k, bound the number of items with each attribute that are allowed to appear in the top k positions of the ranking, the goal is to output a ranking that maximizes the value with respect to the original rank quality metric while respecting the constraints. This problem encapsulates various well-studied problems related to bipartite and hypergraph matching as special cases and turns out to be hard to approximate even with simple constraints. Our main technical contributions are fast exact and approximation algorithms along with complementary hardness results that, together, come close to settling the approximability of this constrained ranking maximization problem. Unlike prior work on the approximability of constrained matching problems, our algorithm runs in linear time, even when the number of constraints is (polynomially) large, its approximation ratio does not depend on the number of constraints, and it produces solutions with small constraint violations. Our results rely on insights about the constrained matching problem when the objective function satisfies certain properties that appear in common ranking metrics such as discounted cumulative gain (DCG), Spearman's rho or Bradley-Terry, along with the nested structure of fairness constraints.
The need for responsible data management intensifies with the growing impact of data on society. One central locus of the societal impact of data are Automated Decision Systems(ADS), socio-legal-technical systems that are used broadly in industry, non-profits, and government. ADS process data about people, help make decisions that are consequential to people’s lives, are designed with the stated goals of improving efficiency and promoting equitable access to opportunity, involve a combination of human and automated decision making, and are subject to auditing for legal compliance and to public disclosure. They may or may not use AI, and may or may not operate with a high degree of autonomy,but they rely heavily on data.In this article, we argue that the data management community is uniquely positioned to lead the responsible design,development, use, and oversight of ADS. We outline a technical research agenda that requires that we step outside our comfort zone of engineering for efficiency and accuracy, to also incorporate reasoning about values and beliefs. This seems high-risk, but one of the upsides is being able to ex-plain to our children what we do and why it matters.
Artificial intelligence (AI) systems built on incomplete or biased data will often exhibit problematic outcomes. Current methods of data analysis, particularly before model development, are costly and not standardized. The Dataset Nutrition Label (the Label) is a diagnostic framework that lowers the barrier to standardized data analysis by providing a distilled yet comprehensive overview of dataset "ingredients" before AI model development. Building a Label that can be applied across domains and data types requires that the framework itself be flexible and adaptable; as such, the Label is comprised of diverse qualitative and quantitative modules generated through multiple statistical and probabilistic modeling backends, but displayed in a standardized format. To demonstrate and advance this concept, we generated and published an open source prototype with seven sample modules on the ProPublica Dollars for Docs dataset. The benefits of the Label are manyfold. For data specialists, the Label will drive more robust data analysis practices, provide an efficient way to select the best dataset for their purposes, and increase the overall quality of AI models as a result of more robust training datasets and the ability to check for issues at the time of model development. For those building and publishing datasets, the Label creates an expectation of explanation, which will drive better data collection practices. We also explore the limitations of the Label, including the challenges of generalizing across diverse datasets, and the risk of using "ground truth" data as a comparison dataset. We discuss ways to move forward given the limitations identified. Lastly, we lay out future directions for the Dataset Nutrition Label project, including research and public policy agendas to further advance consideration of the concept.
Big Data is increasingly mined to rank and rate individuals. Predictive algorithms assess whether we are good credit risks, desirable employees, reliable tenants, valuable customers — or deadbeats, shirkers, menaces, and “wastes of time.” Crucial opportunities are on the line, including the ability to obtain loans, work, housing, and insurance. Though automated scoring is pervasive and consequential, it is also opaque and lacking oversight. In one area where regulation does prevail — credit — the law focuses on credit history, not the derivation of scores from data.
Procedural regularity is essential for those stigmatized by “artificially intelligent” scoring systems. The American due process tradition should inform basic safeguards. Regulators should be able to test scoring systems to ensure their fairness and accuracy. Individuals should be granted meaningful opportunities to challenge adverse decisions based on scores miscategorizing them. Without such protections in place, systems could launder biased and arbitrary data into powerfully stigmatizing scores.
Machine learning model fairness and interpretability are critical for data scientists, researchers and developers to explain their models and understand the value and accuracy of their findings. Interpretability is also important to debug machine learning models and make informed decisions about how to improve them. In this session, Francesca will go over a few methods and tools that enable you to "unpack" machine learning models, gain insights into how and why they produce specific results, assess your AI systems fairness and mitigate any observed fairness issues.
This page last updated 2024-08-27 00:23:19 -0400