The SODALab strives to develop algorithms/markets with provable guarantees to optimize decision-making in dynamic, uncertain, and possibly strategic environments, using mathematical tools from computer science, economics, statistics, and control. In particular, we study the interplay between online algorithms, mechanism design, and learning theory, and its implications for the design and operation of dynamic, multi-agent systems in various fields, ranging from energy and transportation to communication networks. For more details, see some of our recent highlights below.
Online Decision-Making
Research in online decision-making, commonly known as online optimization/learning, focuses on developing online algorithms for a subset of optimization and decision-making problems where (i) data are incrementally revealed over time, and (ii) decisions need to be made sequentially (oftentimes also irrevocably). This differs from offline optimization/learning, where all the data is available upfront, and the algorithm can optimize the solution with complete knowledge of the future. Online algorithms have diverse applications, including online advertising, recommendation systems, financial trading, portfolio management, and supply chain management, etc.
Our research in this thrust has been focusing on (i) competitive online optimization, (ii) no-regret learning, and (iii) beyond worst-case analysis in online decision-making.
Recent publications:
Mechanism Design & Multi-Agent Systems
Mechanism design, a subfield in game theory, combines principles from both economics and computer science. At its core, mechanism design tackles the challenge of designing systems in strategic settings involving multiple self-interested agents. The goal is to develop rules and payoffs that incentivize agents to act in a desired manner (e.g., truthful disclosure of private information), resulting in favorable system-wide outcomes (e.g., welfare maximization). Classical mechanism design in economics often makes distributional assumptions about agent types and overlooks computational constraints. In constrast, the computer science perspective places emphasis on the computational complexity of a designed mechanism and is particularly concerned about whether a mechanism can be efficiently implemented with performance guarantees – usually through the lens of worst-case analysis and approximation ratios. Mechanism design finds broad applications across various domains, including market design, auction theory, network routing, and cloud resource allocation, among others.
Our research in this thrust has been focusing on (i) online/combinatorial auctions, (ii) posted price mechanisms, and (iii) intersection of mechanism design and machine learning.
Recent publications:
Market Design & Energy
Market design plays a critical role in shaping the dynamics and outcomes of various industries, and the energy sector is no exception. As the global demand for reliable, affordable, and sustainable energy continues to rise, the need for effective market structures and mechanisms becomes increasingly evident. Market design in the context of energy involves creating frameworks and rules that facilitate efficient allocation, pricing, and trading of energy resources, while also promoting competition, innovation, and environmental sustainability.
Our research in this thrust has been focusing on developing novel markets for (i) electrical grid with distributed energy resources, (ii) electric vehicle charging, and (iii) the energy-mobility nexus (e.g., electric mobility-on-demand).
