This project aims at designing and implementing a "General Auction Player" (GAP) that can reason about the rules governing an Auction.
Automated agents are now widely used in auction-based markets but state-of-the-art software agents are usually designed to act on a specific market (eg. English auction), and cannot switch between different auctions. To do so, we adopt a perspective imported from "Strategic Reasoning": considering a set of rules describing a market, how its analysis can be embedded into an artificial agent? This agent should be able to "understand" the rules (e.g., how the winner is determined), to reason about her own private information (e.g., what price should she proposed) and also about other players private information (e.g., what she believes about other player's private valuation). Strategic Reasoning abilities are the key elements of GAP definition.
This project considers a logical definition of the players, as logic not only enables to clearly define and reason about strategies and players abilities, but also enables to precisely define the key concepts inherited from Game Theory such as rationality, uncertainty or equilibrium and required by our Strategic Reasoning perspective on auction market.
This project aims at demonstrating that strategic reasoning is actually feasible and can be implemented. To do so, this project considers a lightweight approach by considering as starting point a simple and practical logical language, called Game Description Language (GDL). GDL will then be extended and adapted in order to provide reasoning facilities mixing general principles about strategic reasoning and context-specific knowledge about auction markets.
To reach this goal, we aim at developing a general "Auction Description Language" (ADL), a GDL-based language for representing the rules of an auction. This language will allow a GAP to reason "strategically" in different auction environments.
Developing the agent’s abilities to understand and correctly interpret the rules
underlying different types of auctions. The agent will then be able to reason strategically about
her possible actions, i.e., not only by considering her own goals but also other players’ goals.
The project output will be centered around four main themes:
(i) the development of a logical language based on GDL and dedicated to the representation
of auctions (ADL): this language introduces the core concepts for auction representation
(bidder, auctioneer, shouts...) ;
(ii) an ADL modular specification of auction-based market rules: how to determine the winner,
how to define what bids are admissible...;
(iii) an ADL representation of strategic reasoning, enabling GAP to bid in different types of
auctions: this aspect will then enable players to assess properties such as strategy-proofness;
(iv) an general auctioneer server and GAP prototype.different experiments will be performed in order to assess GAP
general dimension. These sottware will be publicly available and also data about its performancec.
Several lessons will be learned from the completion of project AGAPE. First, the AGAPE project is pioneer in addressing a strategic reasoning problem in concrete terms. This will provide important feedback for the entire community of strategic reasoning on the practical feasibility of logic-based approaches to the practical problem of auctions. AGAPE addresses non-trivial questions about the compact representation of preferences and strategies, and the implementation of reasoning tasks, which will give rise to important contributions to the field of
knowledge representation. On the practical side, by considering multiple kinds of auction of increasing difficulty, the project will assess the balance between the implementation of concepts and principles from strategic reasoning on the one hand, and the benefit of considering
background knowledge about a specific domain on the other hand.
Monsieur Laurent Perrussel (Institut de Recherche en Informatique de Toulouse)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
LAMSADE Laboratoire d'analyse et modélisation de systèmes pour l'aide à la décision
IBISC Informatique, Biologie Intégrative et Systèmes Complexes
LIG Laboratoire d'Informatique de Grenoble
IRIT Institut de Recherche en Informatique de Toulouse
Help of the ANR 349,798 euros
Beginning and duration of the scientific project: January 2019 - 42 Months