Manuel Gomez-Rodriguez, head of the MPI-SWS Human-Centric Machine Learning group, has been awarded an ERC Consolidator Grant for his project "Counterfact: Counterfactuals in Minds and Machines." This project will receive circa 2 million euros over a period of 5 years, starting directly after Manuel's ERC Starting Grant project on "Human-Centered Machine Learning" wraps up next year.

In the most recent round for Consolidator Grants, over 2300 research proposals were submitted to the ERC. The sole selection criterion is scientific excellence. This year, less than 15% of all ERC Consolidator Grant applicants across all scientific disciplines received the award, with only 16 awardees in Computer Science across all of Europe!

Summary of the Counterfact project proposal

Reasoning about what might have been, about alternatives to our own pasts, is a landmark of human intelligence. This type of reasoning, called counterfactual reasoning, is often evaluative, specifying alternatives that are in some sense better or worse than our past reality, and has been shown to play a significant role in the ability that humans have to learn from limited past experience and improve their decision making skills over time. In recent years, there has increasing excitement about the potential of machine learning models and algorithms to support human decision making in a variety of high-stakes domains such as medicine, education
or science. However, these models and algorithms have been traditionally unable to perform, nor benefit from, counterfactual reasoning. In this project, our goal is to bridge this gap.

We will develop machine learning models and algorithms for automated decision support that are able to perform and benefit from counterfactual reasoning in multiple ways. For example, they will perform counterfactual reasoning about human behavior in order to anticipate how humans incorporate algorithmic advice into their decisions. This will enable a new generation of decision support systems that can only increase and never decrease the average quality of human decisions. Moreover, they will use the structural similarities and shared properties across different counterfactual decision making scenarios to significantly reduce their computational and data requirements. In addition, these models and algorithms will also help humans learn from their own past decisions by identifying alternative decisions that would have led to better outcomes. Finally, we will perform large-scale human subject studies with both laypersons and experts to evaluate their effectiveness in a wide variety of decision making tasks.

MPI-SWS doctoral student Eleni Straitouri was awarded a 2024 Google PhD Fellowship (in Machine Intelligence) to fund her doctoral research on using machine learning to build automated decision support systems that can reliably improve the decisions of human experts.

The Google PhD Fellowship Program was created to recognize outstanding graduate students doing exceptional and innovative research in areas relevant to computer science and related fields. Eleni, who is advised by Manuel Gomez Rodriguez, was one of only 85 recipients worldwide in 2024.

Link: https://research.google/outreach/phd-fellowship/

Mariya Toneva joins the tenure-track faculty at our institute starting September 2022. Mariya’s research is at the intersection of machine learning, natural language processing, and neuroscience. Her group bridges language in machines with language in the brain, with a focus on building computational models of language processing in the brain that can also improve natural language processing systems.

Prior to joining MPI-SWS, Mariya conducted research as a C.V. Starr Fellow at the Princeton Neuroscience Institute. She received her Ph.D. in a joint program between Machine Learning and Neural Computation from Carnegie Mellon University, and her B.S. in Computer Science and Cognitive Science from Yale University.

Two MPI-SWS faculty, Maria Christakis and Elissa Redmiles, have earned highly competitive Google Research Scholar awards. Maria Christakis's award was given for her research on metamorphic specification and testing of machine-learning models and Elissa Redmiles's award was given for her research on aligning technical data privacy protections with user concerns.

The Google Research Scholar Program provides unrestricted gifts of up to $60,000 to support research at institutions around the world and is focused on funding world-class research conducted by early-career professors. Award proposals go through an internal, merit-based review process and selected faculty can receive a Google Research Scholar award only once in their career. Award recipients are assigned a liaison at the company to share findings with and as a point of contact for further collaboration.

MPI-SWS researchers Jiarui Gan, Rupak Majumdar, Goran Radanovic, and Adish Singla have received an Outstanding Paper Award Honorable Mention at AAAI 2022, for their paper "Bayesian Persuasion in Sequential Decision-Making."

The AAAI conference is one of the leading international venues for AI research, covering all sub-areas of the field. AAAI 2022 received more than 9000 submissions, of which 1370 were accepted for publication. Of these 1370 papers, only three papers were selected for an outstanding paper award. Of these three outstanding papers, two were authored by researchers at the Saarland Informatics Campus.

Adish Singla, head of the MPI-SWS Machine Teaching Research group, has been awarded a 2021 ERC Starting Grant. Over the next five years, his project TOPS will receive funding of 1.5 million euros for research on "Machine-Assisted Teaching for Open-Ended Problem Solving: Foundations and Applications". Read more about the TOPS project below.

ERC grants are the most prestigious and the most competitive European-level awards for ground-breaking scientific investigations. This year, less than 10% of all ERC Starting Grant applicants across all scientific disciplines received the award, with only 23 awardees in Computer Science across all of Europe! You can find more information about ERC Starting Grants awarded this year at https://erc.europa.eu/news/StG-recipients-2021.

The TOPS Project

Computational thinking and problem solving skills are essential for everyone in the 21st century, both for students to excel in STEM+Computing fields and for adults to thrive in the digital economy. Consequently, educators are putting increasing emphasis on pedagogical tasks in open-ended domains such as programming, conceptual puzzles, and virtual reality environments.

When learning to solve such open-ended tasks by themselves, people often struggle. The difficulties are embodied in the very nature of tasks being open-ended: (a) underspecified (multiple solutions of variable quality), (b) conceptual (no well-defined procedure), (c) sequential (series of interdependent steps needed), and (d) exploratory (multiple pathways to reach a solution). These struggling learners can benefit from individualized assistance, for instance, by receiving personalized curriculum across tasks or feedback within a task. Unfortunately, human tutoring resources are scarce, and receiving individualized human-assistance is rather a privilege. Technology empowered by artificial intelligence has the potential to tackle this scarcity challenge by providing scalable and automated machine-assisted teaching. However, the state-of-the-art technology is limited: it is designed for well-defined procedural learning, but not for open-ended conceptual problem solving.

The TOPS project will develop next-generation technology for machine-assisted teaching in open-ended domains. We will design novel algorithms for assisting the learner by bridging reinforcement learning, imitation learning, cognitive science, and symbolic reasoning. Our theoretical foundations will be based on a computational framework that models the learner as a reinforcement learning agent who gains mastery with the assistance of an automated teacher. In addition to providing solid foundations, we will demonstrate the performance of our techniques in a wide range of pedagogical applications.

Sumit Gulwani, a scientist at Microsoft Research in Redmond, is one of two recipients of the 2021 Max Planck-Humboldt Medal. He will collaborate with MPI-SWS faculty Rupak Majumdar and Adish Singla on the problem of AI in Education.

With a background in program analysis and artificial intelligence, Sumit Gulwani shaped the field of program synthesis, which emerged around 2010. The computer scientist developed algorithms that can efficiently generate computer programs from very few input-output examples, natural-language-based specification, or from just the code and data context. His work made it possible for non-programmers to program tedious, repetitive spreadsheet tasks, and enabled productivity improvements for data scientists and developers for data wrangling and software engineering tasks. Recently, Sumit has also been using the tools of program synthesis for computer-aided education of pupils and students. Starting from the automatic correction of learners' work in programming education, he further evolved this line of work to detect misunderstandings and give learning feedback and grades, also in subjects like mathematics and language learning.

The medal comes with prize money in the amount of 60,000 euros.

MPI-SWS faculty member Elissa Redmiles, along with collaborators Samuel Dooley and Professor John Dickerson from the University of Maryland as well as Professor Dana Turjeman from Reichman University, helped the Louisiana Department of Health advertise their COVID19 contact tracing app. As part of this work, the researchers conducted a randomized, controlled field experiment to provide guidance to other jurisdictions on how to most effectively and ethically advertise these public health tools.The Linux Public Health Foundation has featured their findings as guidance for other jurisdictions looking to advertise their contact tracing apps.

This work is part of a larger project Redmiles leads on ethical adoption of COVID 19 apps: https://covidadoptionproject.mpi-sws.org/.

MPI-SWS faculty member Elissa Redmiles was quoted in Rolling Stone, El Pais -- Spain's largest newspaper -- and in NetzPolitik regarding ongoing research in collaboration with MPI-SWS group member Vaughn Hamilton and MPI-SWS Intern Hanna Barakat (also at Brown University) on the shift toward digital sex work as a result of COVID19, as well as her work with collaborators Catherine Barwulor (Clemson University), Allison McDonald (University of Michigan), and Eszter Hargittai (University of Zurich) on digital discrimination against European sex workers, originally published in ACM CHI.

Ralf Jung and Bilal Zafar have each been awarded a 2021 Otto Hahn Medal for outstanding scientific achievement. The Max Planck Society awards the Otto Hahn Medal annually to young scientists in recognition of outstanding scientific achievement. Ralf was awarded the medal for his work on the first formal foundations for the cutting-edge systems programming language Rust, while Bilal was awarded the medal for his work on developing responsible and trustworthy AI systems that can help reduce discrimination and polarisation in society. Ralf obtained his PhD in August 2020, and was advised by Derek Dreyer. Ralf is now a postdoc at MPI-SWS and research affiliate at MIT. Bilal obtained his PhD in February 2019, and was advised by Krishna Gummadi and Manuel Gomez Rodriguez. Bilal is now an Applied Scientist at Amazon Web Services.

Ever since the birth of Artificial Intelligence (AI) at the Dartmouth workshop in 1956, researchers have debated about the exact role that AI will play, and should play, in society. While some have envisioned a romanticized version of AI, incorporated into the narratives of 20th century movies, successful AI developments are often closer to J. C. R. Licklider’s vision of AI, which puts an emphasis on a collaborative relationship between humans and AI, and focuses on hybrid human-AI decision making.

In the Multi-Agent Systems group at MPI-SWS, we study multi-agent sequential decision making using formal frameworks that can capture nuances often presented in human-AI collaborative settings. Specifically, we study different aspects of agent-to-agent interaction in settings where agents share a common goal, but can have different perceptions of reality. The overall goal is to design a more effective AI decision maker that accounts for the behavior of its collaborators, and compensates for their imperfections. To achieve this goal, the AI decision maker can use steering policies to nudge its collaborators to adopt better policies, i.e., policies that lead to an improved joint outcome. In what follows, we summarize some of our recent results related to this agenda.

Accounting for misaligned world-views. An effective way to model behavioral differences between humans and modern AI tools (based on machine learning) is through a model that captures the misalignment in how the agents perceive their environment. Using this approach, we have proposed a new computational model, called Multi-View Decision Process, suitable for modeling two-agent cooperative scenarios in which agents agree on their goals, but disagree on how their actions affect the state of the world [1]. This framework enables us to formally analyze the utility of accounting for the misalignment in agents’ world-views when only one of the agents has a correct model of the world. Our results show that modeling such a misalignment is not only beneficial, but critical. The main takeaway is that to facilitate a more successful collaboration among agents, it is not sufficient to make one agent (more) accurate in its world-view: naively improving the accuracy of one agent can degrade the joint performance unless one explicitly accounts for the imperfections of the other agent. To this end, we have developed an algorithm for finding an approximately optimal steering policy for the agent with the correct world-view.

Adapting to a non-stationary collaborator. In addition to accounting for a misalignment in world-views, decision makers must also account for the effects of their behavior on other agents. Namely, decision makers respond to each other's behavior, leading to behavior which is non-stationary and changes over time. In the context of human-AI collaboration, this might happen if the human agent changes their behavior over time, for example, as it learns to interact with the AI agent. Such non-stationary behavior of the human agent could have a negative impact on the collaboration, and can lead to a substantially worse performance unless the AI agent adapts to the changing behavior of the human agent. We can model this situation with a two-agent setting similar to the one presented above, but which allows agents to change their behavior as they interact over time [2]. The agent with the correct world-view now has to adapt to the non-stationary behavior of its collaborator. We have proposed a learning procedure that has provable guarantees on the joint performance under the assumption that the behavior of the other agent is not abruptly changing over time. We have shown that this assumption is not trivial to relax in that obtaining the same guarantees without this assumption would require solving a computationally intractable problem.

Steering via environment design. The previous two cases consider indirect steering policies for which the agent with the correct model implicitly influences the behavior of its collaborator by acting in the world. A more explicit influence would be obtained if the actions of this agent are directly changing the world-view of its collaborator. In the context of human-AI collaboration, the AI agent could shape the environment to nudge the human agent to adopt a more efficient decision policy. This can be done through reward shaping, i.e., by making some actions more costly for humans in terms of effort, or through dynamics shaping, i.e., by changing the perceived influence that the human’s actions have on the world. In the machine learning terminology, such a steering strategy is nothing else but a form of an adversarial attack of the AI agent (attacker) on the human agent. In our recent work [3], we have characterized how to optimally perform these types of attacks and how costly they are from an attacker’s point of view.

 

References: 

[1] Dimitrakakis, C., Parkes, D.C., Radanovic, G. and Tylkin, P., 2017. Multi-view Decision Processes: The Helper-AI Problem. In Advances in Neural Information Processing Systems.

[2] Radanovic, G., Devidze, R., Parkes, D. and Singla, A., 2019. Learning to Collaborate in Markov Decision Processes. In International Conference on Machine Learning.

[3] Rakhsha, A., Radanovic, G., Devidze, R., Zhu, X. and Singla, A., 2020. Policy Teaching via Environment Poisoning: Training-time Adversarial Attacks against Reinforcement Learning. In International Conference on Machine Learning.

Manuel Gomez-Rodriguez, head of the MPI-SWS Human-Centric Machine Learning group, has been awarded an ERC Starting Grant. Over the next five years, his project "Human-Centric Machine Learning" will receive 1.49 million euros, which will allow the group to develop the foundations of human-centric machine learning.

In the most recent round for Starting Grants, over 3300 research proposals were submitted to the ERC. The sole selection criterion is scientific excellence. This year, less than 14% of all ERC Starting Grant applicants across all scientific disciplines received the award, with only 20 awardees in Computer Science across all of Europe!

Summary of the HumanML project proposal

With the advent of mass-scale digitization of information and virtually limitless computational power, an increasing number of social, information and cyber-physical systems evaluate, support or even replace human decisions using machine learning models and algorithms. Machine learning models and algorithms have been traditionally designed to take decisions autonomously, without human intervention, on the basis of passively collected data. However, in most social, information and cyber-physical systems, algorithmic and human decisions feed on and influence each other. As these decisions become more consequential to individuals and society, machine learning models and algorithms have been blamed for playing a major role in an increasing number of missteps, from discriminating against minorities, causing car accidents and increasing polarization to misleading people in social media.

In this project, we will develop human-centric machine learning models and algorithms for evaluating, supporting and enhancing decision-making processes where algorithmic and human decisions feed on and influence each other. These models and algorithms will account for the feedback loop between algorithmic and human decisions, which currently perpetuates or even amplifies biases and inequalities, and they will learn to operate under different automation levels. Moreover, they will anticipate how individuals will react to their algorithmic decisions, often strategically, to receive beneficial decisions and they will provide actionable insights about their algorithmic decisions. Finally, we will perform observational and interventional experiments as well as realistic simulations to evaluate their effectiveness in a wide range of applications, from content moderation, recidivism prediction, and credit scoring to medical diagnosis and autonomous driving.

Goran Radanovic joined MPI-SWS as a research group leader on Sep 16, 2019. He is generally interested in studying AI systems, and more specifically in the design and analysis of systems with intelligent and self-interested agents. Particular topics of interest include value-aligned artificial intelligence, human-AI collaboration, and decision making systems with societally-aware utility functions. 

Prior to joining MPI-SWS, he was a postdoctoral researcher at Harvard University, where he worked with Prof. David C. Parkes. He received his Ph.D. in Computer Science from the Swiss Federal Institute of Technolgy in Lausanne (EPFL), under the supervision of Prof. Boi Faltings. He obtained his Master’s and Bachelor’s degrees in Computer Science fromthe Faculty of Electrical Engineering and Computing at the University of Zagreb.

The Max Planck Institute for Software Systems (MPI-SWS) is inviting applications for Junior Research Group leader positions in systems and machine learning (SysML), human-oriented machine learning (fairness, accountability, transparency, and ethical aspects of AI), adversarial ML, reinforcement learning,  human-computer interaction with ML/social aspects, natural language processing, and learning & cognitive sciences.

Our Junior Research Group program offers young scientists the opportunity to develop their own research program. The position is funded for 5 years with the possibility of a 2-year extension. Applicants must have completed a doctoral degree in computer science or related areas and must have demonstrated outstanding research vision and potential to successfully lead a research group. Successful candidates are expected to build a highly visible research agenda, to mentor junior scientists, and to participate in collaborative projects.

The Max Planck Institute for Software Systems is located in Saarbruecken and Kaiserslautern in Germany. We maintain an open, international, and diverse work environment and seek applications from outstanding researchers regardless of national origin. Our working language is English. We collaborate with several major research institutions worldwide and have high international visibility. There is generous travel, administrative, and technical support available for all group members.

Please apply at https://apply.mpi-sws.org/ under ``Research Group Leader''. You need to upload your CV, a research plan, an optional teaching statement, and 3-5 references. Reviewing of applications will commence on 15 May 2019 and will continue until the positions are filled. The expecting starting date for the position is Fall 2019. Informal inquiries can be addressed to applications-sis@mpi-sws.org.

The Max Planck Society is committed to employing more individuals with disabilities and expressly welcomes them to apply. The Max Planck Society seeks to increase the percentage of women in the areas where they are underrepresented and expressly welcomes them to apply.

MPI-SWS faculty member Krishna Gummadi and MPI-SWS alumnus Alan Mislove have been awarded a "Secure the Internet" grant by Facebook. Their proposal, “Towards privacy-protecting aggregate statistics in PII-based targeted advertising,” has been awarded $60,000 to develop techniques for revealing advertising statistics that provide hard guarantees of user privacy, based on a (principles-first) approach. Their goal is to develop a differential privacy-like approach that can be applied to existing advertising systems.

The Facebook "Secure the Internet" grant program is designed to improve the security, privacy, and safety of internet users. Gummadi and Mislove's proposal was one of only 10 winning proposals, which were together awarded more than $800,000 by Facebook.

Many real-world systems involve repeatedly making decisions under uncertainty—for instance, choosing one of the several products to recommend to a user in an online recommendation service, or dynamically allocating resources among available stock options in a financial market. Machine learning (ML) algorithms driving these systems typically operate under the assumption that they are interacting with static components, e.g., users' preferences are fixed, trading tools providing stock recommendations are static, and data distributions are stationary. This assumption is often violated in modern systems, as these algorithms are increasingly interacting with and seeking information from learning agents including people, robots, and adaptive adversaries. Consequently, many well-studied ML frameworks and algorithmic techniques fail to provide desirable theoretical guarantees—for instance, algorithms might converge to a sub-optimal solution or fail arbitrarily bad in these settings.

Researchers at the Machine Teaching Group, MPI-SWS are designing novel ML algorithms that have to interact with agents that are adaptive or learning over time, especially in situations when the algorithm's decisions directly affect the state dynamics of these agents. In recent work [1], they have studied the above-mentioned problem in the context of two fundamental machine learning frameworks: (i) online learning using experts' advice and (ii) active learning using labeling oracles. In particular, they consider a setting where experts/oracles themselves are learning agents. For instance, active learning algorithms typically query labels from an oracle, e.g., a (possibly noisy) domain expert; however, in emerging crowd-powered systems, these experts are getting replaced by inexpert participants who could themselves be learning over time (e.g., volunteers in citizen science projects). They have shown that when these experts/oracles themselves are learning agents, well-studied algorithms (like the EXP3 algorithm) fail to converge to the optimal solution and can have arbitrarily bad performance for this new problem setting. Furthermore, they provide an impossibility result showing that without sharing any information across experts, it is impossible to achieve convergence guarantees. This calls for developing novel algorithms with practical ways of coordination between the central algorithm and learning agents to achieve desired guarantees.

Currently, researchers at the Machine Teaching Group are studying these challenges in the context of designing next-generation human-AI collaborative systems. As a concrete application setting, consider a car driving scenario where the goal is to develop an assistive AI agent to drive the car in an auto-pilot mode, but giving control back to the human driver in safety-critical situations. They study this setting by casting it as a multi-agent reinforcement learning problem. When the human agent has a stationary policy (i.e., the actions take by the human driver in different states/scenarios are fixed), it is trivial to learn an optimal policy for the AI agent that maximizes the overall performance of this collaborative system. However, in real-life settings where a human driver would adapt their behavior in response to the presence of an auto-pilot mode, they show that the problem of learning an optimal policy for the AI agent becomes computationally intractable. This work is one of the recent additions to an expanding set of results and algorithmic techniques developed by MPI-SWS researchers in the nascent area of Machine Teaching [2, 3].

References

[1] Adish Singla, Hamed Hassani, and Andreas Krause. Learning to Interact with Learning Agents. In Proceedings of the 32nd AAAI Conference on Artificial Intelligence (AAAI'18), 2018.

[2] Xiaojin Zhu, Adish Singla, Sandra Zilles, and Anna N. Rafferty. An Overview of Machine Teaching. arXiv 1801.05927, 2018.

[3] Maya Cakmak, Anna N. Rafferty, Adish Singla, Xiaojin Zhu, and Sandra Zilles. Workshop on Teaching Machines, Robots, and Humans. NIPS 2017.

Krishna Gummadi, head of the MPI-SWS Networked Systems group, has been awarded an ERC Advanced Grant. Over the next five years, his project "Foundations of Fair Social Computing" will receive 2.49 million euros, which will allow the group to develop the foundations for fair social computing in the future.

In the most recent round for Advanced Grants, a total of 2,167 research proposals were submitted to the ERC out of which merely 12% were selected for funding. The sole selection criterion is scientific excellence.

Summary of the Fair Social Computing project proposal

Social computing represents a societal-scale symbiosis of humans and computational systems, where humans interact via and with computers, actively providing inputs to influence---and in turn being influenced by---the outputs of the computations. Social computations impact all aspects of our social lives, from what news we get to see and who we meet to what goods and services are offered at what price and how our creditworthiness and welfare benefits are assessed. Given the pervasiveness and impact of social computations, it is imperative that social computations be fair, i.e., perceived as just by the participants subject to the computation. The case for fair computations in democratic societies is self-evident: when computations are deemed unjust, their outcomes will be rejected and they will eventually lose their participants.

Recently, however, several concerns have been raised about the unfairness of social computations pervading our lives, including

1. the existence of implicit biases in online search and recommendations,


2. the potential for discrimination in machine learning based predictive analytics, and


3. a lack of transparency in algorithmic decision making, with systems providing little to no information about which sensitive user data they use or how they use them.

Given these concerns, we need reliable ways to assess and ensure the fairness of social computations. However, it is currently not clear how to determine whether a social computation is fair, how we can compare the fairness of two alternative computations, how to adjust a computational method to make it more fair, or how to construct a fair method by design. This project will tackle these challenges in turn. We propose a set of comprehensive fairness principles, and will show how to apply them to social computations. In particular, we will operationalize fairness, so that it can be measured from empirical observations. We will show how to characterize which fairness criteria are satisfied by a deployed computational system. Finally, we will show how to synthesize non-discriminatory computations, i.e., how to learn an algorithm from training data that satisfies a given fairness principle.

Machine learning algorithms are increasingly being used to automate decision making in several domains such as hiring, lending and crime-risk prediction. These algorithms have shown significant promise in leveraging large or “big” training datasets to achieve high prediction accuracy, sometimes surpassing even human accuracy.

Unfortunately, some recent investigations have shown that machine learning algorithms can also lead to unfair outcomes. For example, a recent ProPublica study found that COMPAS, a tool used in US courtrooms for assisting judges with crime risk prediction, was unfair towards black defendants. In fact, several studies from governments, regulatory authorities, researchers as well as civil rights groups have raised concerns about machine learning potentially acting as a tool for perpetuating existing unfair practices in society, and worse, introducing new kinds of unfairness in prediction tasks. As a consequence, a flurry of recent research has focused on defining and implementing appropriate computational notions of fairness for machine learning algorithms.

Parity-based fairness

Existing computational notions of fairness in the machine learning literature are largely inspired by the concept of discrimination in social sciences and law. These notions require the decision outcomes to ensure parity (i.e. equality) in treatment and in impact.

Notions based on parity in treatment require that the decision algorithm should not take into account the sensitive feature information (e.g., gender, race) of a user. Notions based on parity in impact require that the decision algorithm should give beneficial decision outcomes (e.g., granting a loan) to similar percentages of people from all sensitive feature groups (e.g., men, women).

However, in many cases, these existing notions are too stringent and can lead to unexpected side effects. For example, ensuring parity has been shown to lead to significant reductions in prediction accuracy. Parity may also lead to scenarios where none of the groups involved in decision making (e.g., neither men nor women) get beneficial outcomes. In other words, these scenarios might be preferred neither by the decision maker using the algorithm (due to diminished accuracy), nor by the groups involved (due to very little benefits).

User preferences and fairness

In recent work, to appear at NIPS 2017, researchers at MPI-SWS have introduced two new computational notions of algorithmic fairness: preferred treatment and preferred impact. These notions are inspired by ideas related to envy-freeness and bargaining problem in economics and game theory. Preferred treatment and preferred impact leverage these ideas to build more accurate solutions that are preferable for both the decision maker and the user groups.

The new notion of preferred treatment allows basing the decisions on sensitive feature information (thereby relaxing the parity treatment criterion) as long as the decision outcomes do not lead to envy. That is, each group of users prefers their own group membership over other groups and does not feel that presenting itself to the algorithm as another group would have led to better outcomes for the group.

The new notion of preferred impact allows differences in beneficial outcome rates for different groups (thereby relaxing the parity impact criterion) as long as all the groups get more beneficial outcomes than what they would have received under the parity impact criterion.

In their work, MPI-SWS researchers have developed a technique to ensure machine learning algorithms satisfy preferred treatment and / or preferred impact. They also tested their technique by designing crime-predicting machine-learning algorithms that satisfy the above-mentioned notions. In their experiments, they show that preference-based fairness notions can provide significant gains in overall decision-making accuracy as compared to parity-based fairness, while simultaneously increasing the beneficial outcomes for the groups involved.

This work is one of the most recent additions to an expanding set of techniques developed by MPI-SWS researchers to enable fairness, accountability and interpretability of machine learning algorithms.

References

Bilal Zafar, Isabel Valera, Manuel Gomez Rodriguez, Krishna Gummadi and Adrian Weller. From Parity to Preference: Learning with Cost-effective Notions of Fairness. Neural Information Processing Systems (NIPS), Long Beach (CA, USA), December 2017

MPI-SWS researchers—faculty members Krishna Gummadi and Peter Druschel and former SWS doctoral students Alan Mislove and Massimiliano Marcon—have received the ACM SIGCOMM Test of Time Award for their IMC 2007 paper on "Measurement and Analysis of Online Social Networks." The work was done in collaboration with Bobby Bhattacharjee of the University of Maryland.

The award citation reads as follows: "This is one of the first papers that examine multiple online social networks at scale. By introducing novel measurement techniques, the paper has had an enduring influence on the analysis, modeling and design of modern social media and social networking services."

We are pleased to announce that two new faculty will join MPI-SWS.



Allen Clement obtained his Ph.D. at the University of Texas at Austin in 2011. Allen's research aims at designing and building systems that continue to work despite the myriad of things that go 'wrong' in deployed systems, including broken components, malicious adversaries, and benign race conditions. His research builds on techniques from distributed systems, security, fault tolerance, and game theory.



Cristian Danescu-Niculescu-Mizil is joining us from Cornell University, where he obtained his PhD in computer science. Cristian's research aims at developing computational frameworks that can lead to a better understanding of human social behavior, by unlocking the unprecedented potential of the large amounts of natural language data generated online. His work tackles problems related to conversational behavior, opinion mining, computational semantics and computational advertising.

A study by MPI-SWS researchers Saikat Guha (now at Microsoft Research), Bin Cheng, and Paul Francis has been highlighted on CNN, NPR, The Washington Post, Fox News, and other major media outlets.

The study, which will be presented at the ACM Internet Measurement Conference (IMC) in November, looks at the targeting behavior of Google and Facebook. While the goal of the study was to understand targeting in general, the researchers discovered that gay Facebook users can unknowingly reveal to advertisers that they are gay simply by clicking on an ad targeted to gay men. The ads appear innocuous in that they make no mention of targeting gay users (for instance, an ad for a nursing degree). A user's sexual orientation can be leaked even if the user made his sexual orientation private using Facebook's privacy settings.

This study was done as part of a broader research project to design techniques for making advertising more private.