Periodic Reporting for period 4 - MIAMI (Machine Learning-based Market Design)
Berichtszeitraum: 2023-06-01 bis 2023-11-30
Market designers study how to set the "rules of a marketplace" such that the market works well. This is important for society, such that resources are allocated efficiently (put to good use), such that market participants do not have an incentive to manipulate the market mechanism, and such that the marketplace produces a fair outcome.
However, markets are getting increasingly complex such that designing good market mechanisms "by hand" is often infeasible, in particular when certain design desiderata are in conflict with each other. Moreover, humans are boundedly-rational: already in small domains, they are best modeled as having incomplete preferences, because they may only know a ranking or the values of their top choices. In combinatorial domains (where participants make choices regarding bundles of options), the number of choices grows exponentially, such that it quickly becomes impossible for humans to report their full preferences to the market mechanism.
In this ERC project, we combine techniques from "machine learning" with "market design" to address these challenges. The main idea is that machine learning can help deal with incomplete data. For example, a machine learning algorithm can learn the preferences of a human in a marketplace from just a few observations, and these learned preference functions can then be used by a market mechanism to make better decisions. But this also introduces new challenges, as the learned preferences may be wrong, or the market participants may manipulate the machine learning algorithm on purpose. Thus, when designing machine learning-based market mechanism, we take these challenges into account.
In addition to pushing the scientific boundaries of market design research, this ERC project also has an immediate impact on practical market design. We apply our techniques in two different settings: (1) for the design of combinatorial spectrum auctions, a multi-billion dollar domain; and (2) for the design of matching markets (e.g. school choice, refugee matching, adoption matching, course allocation).
However, markets are getting increasingly complex such that designing good market mechanisms "by hand" is often infeasible, in particular when certain design desiderata are in conflict with each other. Moreover, humans are boundedly-rational: already in small domains, they are best modeled as having incomplete preferences, because they may only know a ranking or the values of their top choices. In combinatorial domains (where participants make choices regarding bundles of options), the number of choices grows exponentially, such that it quickly becomes impossible for humans to report their full preferences to the market mechanism.
In this ERC project, we combine techniques from "machine learning" with "market design" to address these challenges. The main idea is that machine learning can help deal with incomplete data. For example, a machine learning algorithm can learn the preferences of a human in a marketplace from just a few observations, and these learned preference functions can then be used by a market mechanism to make better decisions. But this also introduces new challenges, as the learned preferences may be wrong, or the market participants may manipulate the machine learning algorithm on purpose. Thus, when designing machine learning-based market mechanism, we take these challenges into account.
In addition to pushing the scientific boundaries of market design research, this ERC project also has an immediate impact on practical market design. We apply our techniques in two different settings: (1) for the design of combinatorial spectrum auctions, a multi-billion dollar domain; and (2) for the design of matching markets (e.g. school choice, refugee matching, adoption matching, course allocation).
We have written eight papers towards the foundations of preference learning and towards automatically learning new mechanisms, with a recent focus on dynamic settings:
1. "NOMU: Neural Optimization-based Model Uncertainty" introduces a new ML framework to assess neural network prediction uncertainty, improving market mechanism preference elicitation. Published at ICML'22.
2. "Fourier Analysis-based Iterative Combinatorial Auctions" combines Fourier analysis with deep learning to uncover hidden structures in agents’ preferences, enhancing learning speed. Published at IJCAI'22.
3. "Monotone-Value Neural Networks" utilizes monotonicity in agents' preferences to enhance combinatorial assignment domain learning, boosting market mechanism efficiency. Published at IJCAI'22.
4. "Differentiable Economics for Randomized Affine Maximizer Auctions" explores neural network-driven automated mechanism design, presenting a novel architecture for strategyproof lotteries. Published at IJCAI'23.
5. "Learning Solutions in Large Economic Networks using Deep Multi-Agent Reinforcement Learning" demonstrates finding solutions in complex networks with many agents using structured learning. Published at AAMAS'23.
6. "Automated Design of Affine Maximizer Mechanisms In Dynamic Settings" extends affine maximizer mechanisms to dynamic environments, addressing untruthful reward reporting through bilevel optimization. Published at AAAI'24.
7. "Computing Perfect Bayesian Equilibria in Sequential Auctions with Verification" proposes an algorithm for computing equilibria in auctions with a verification phase to limit utility loss. Under review at IJCAI'24.
8. "Learning Best Response Policies in Dynamic Auctions" focuses on optimizing bidding strategies in dynamic auctions through a Markov Decision Process and multi-task reinforcement learning. Preparing for AAAI'25 submission.
We have written 10 papers, studying the question how to best integrate a machine learning algorithm into a market mechanism to increase its efficiency:
1. "Deep Learning-powered Iterative Combinatorial Auctions" introduces the first deep learning-driven combinatorial auction, improving efficiency and runtime. Published at AAAI 2020.
2. "iMLCA: Machine Learning-powered Iterative Combinatorial Auctions with Interval Bidding" presents a ML-powered auction reducing bidders' cognitive effort. Published at ACM EC 2021.
3. "Market Design for Drone Traffic Management" opens research on drone traffic management, identifying key challenges and solutions. Published at AAAI’22.
4. "Bayesian Optimization-based Combinatorial Assignment" integrates model uncertainty into combinatorial auctions. Published at AAAI’23.
5. "Machine Learning-powered Clock Auctions" develops an ML-based auction using demand queries to clear the market efficiently. Published at AAA’24.
6. "Scalable Mechanism Design for Multi-Agent Path Finding" applies mechanism design to drone traffic management. Under review at IJCAI’24 submission.
7. "Machine Learning-powered Course Allocation" improves efficiency and fairness in university course allocation using ML for preference elicitation. Under review at EC’24.
8. "Search and Matching for Adoption from Foster Care" compares traditional and ML-based adoption matching platforms. Under review at EC’24.
9. "Machine Learning-powered Iterative Combinatorial Auctions" employs support vector regression in auction design, aiming for efficiency and good incentive properties. Preparing for journal submission.
10. "Enabling Trade-offs in Machine Learning-based Matching for Refugee Resettlement" develops an ML algorithm for refugee allocation balancing interests.
1. "NOMU: Neural Optimization-based Model Uncertainty" introduces a new ML framework to assess neural network prediction uncertainty, improving market mechanism preference elicitation. Published at ICML'22.
2. "Fourier Analysis-based Iterative Combinatorial Auctions" combines Fourier analysis with deep learning to uncover hidden structures in agents’ preferences, enhancing learning speed. Published at IJCAI'22.
3. "Monotone-Value Neural Networks" utilizes monotonicity in agents' preferences to enhance combinatorial assignment domain learning, boosting market mechanism efficiency. Published at IJCAI'22.
4. "Differentiable Economics for Randomized Affine Maximizer Auctions" explores neural network-driven automated mechanism design, presenting a novel architecture for strategyproof lotteries. Published at IJCAI'23.
5. "Learning Solutions in Large Economic Networks using Deep Multi-Agent Reinforcement Learning" demonstrates finding solutions in complex networks with many agents using structured learning. Published at AAMAS'23.
6. "Automated Design of Affine Maximizer Mechanisms In Dynamic Settings" extends affine maximizer mechanisms to dynamic environments, addressing untruthful reward reporting through bilevel optimization. Published at AAAI'24.
7. "Computing Perfect Bayesian Equilibria in Sequential Auctions with Verification" proposes an algorithm for computing equilibria in auctions with a verification phase to limit utility loss. Under review at IJCAI'24.
8. "Learning Best Response Policies in Dynamic Auctions" focuses on optimizing bidding strategies in dynamic auctions through a Markov Decision Process and multi-task reinforcement learning. Preparing for AAAI'25 submission.
We have written 10 papers, studying the question how to best integrate a machine learning algorithm into a market mechanism to increase its efficiency:
1. "Deep Learning-powered Iterative Combinatorial Auctions" introduces the first deep learning-driven combinatorial auction, improving efficiency and runtime. Published at AAAI 2020.
2. "iMLCA: Machine Learning-powered Iterative Combinatorial Auctions with Interval Bidding" presents a ML-powered auction reducing bidders' cognitive effort. Published at ACM EC 2021.
3. "Market Design for Drone Traffic Management" opens research on drone traffic management, identifying key challenges and solutions. Published at AAAI’22.
4. "Bayesian Optimization-based Combinatorial Assignment" integrates model uncertainty into combinatorial auctions. Published at AAAI’23.
5. "Machine Learning-powered Clock Auctions" develops an ML-based auction using demand queries to clear the market efficiently. Published at AAA’24.
6. "Scalable Mechanism Design for Multi-Agent Path Finding" applies mechanism design to drone traffic management. Under review at IJCAI’24 submission.
7. "Machine Learning-powered Course Allocation" improves efficiency and fairness in university course allocation using ML for preference elicitation. Under review at EC’24.
8. "Search and Matching for Adoption from Foster Care" compares traditional and ML-based adoption matching platforms. Under review at EC’24.
9. "Machine Learning-powered Iterative Combinatorial Auctions" employs support vector regression in auction design, aiming for efficiency and good incentive properties. Preparing for journal submission.
10. "Enabling Trade-offs in Machine Learning-based Matching for Refugee Resettlement" develops an ML algorithm for refugee allocation balancing interests.
I'd like to highlight three areas where we have made significant progress beyond the state of the art:
1. Our work on machine learning-powered combinatorial auctions has shown, for the first time, that incorporating machine learning algorithms into an iterative combinatorial auction can lead to significant efficiency improvements. In large domains in particular, our designs outperform the combinatorial clock auction (CCA), the auction design most often used in practice. Using the "uncertainty quantification" for neural networks we have developed, we were able to develop Bayesian Optimization technique that can be incorporated into an iterative combinatorial auction, which has further increased the efficiency of the mechanism.
2. In our work on matching markets, we have developed the first model to study adoption matching mechanisms from a market design perspective. This allowed us to compare two of the main approaches how adoption matchings are currently performed regarding their social welfare.
3. In our work on course allocation, we have developed a new machine learning-based course allocation mechanism, which outperforms the current state-of-the-art mechanism (Course Match) in terms of allocative efficiency and fairness. In this project in particular, we have shown that ameliorating agents' reporting mistakes can be more important for efficiency and fairness than the choice of the mechanism itself.
1. Our work on machine learning-powered combinatorial auctions has shown, for the first time, that incorporating machine learning algorithms into an iterative combinatorial auction can lead to significant efficiency improvements. In large domains in particular, our designs outperform the combinatorial clock auction (CCA), the auction design most often used in practice. Using the "uncertainty quantification" for neural networks we have developed, we were able to develop Bayesian Optimization technique that can be incorporated into an iterative combinatorial auction, which has further increased the efficiency of the mechanism.
2. In our work on matching markets, we have developed the first model to study adoption matching mechanisms from a market design perspective. This allowed us to compare two of the main approaches how adoption matchings are currently performed regarding their social welfare.
3. In our work on course allocation, we have developed a new machine learning-based course allocation mechanism, which outperforms the current state-of-the-art mechanism (Course Match) in terms of allocative efficiency and fairness. In this project in particular, we have shown that ameliorating agents' reporting mistakes can be more important for efficiency and fairness than the choice of the mechanism itself.