Each group of students will implement a trading agent to compete in an Ad-Auction simulation. Specifically, we will use the server infrastructure of the Trading Agents Competition on Ad Auctions to simulate populations of users and a publisher ad auction platform. The trading agents by the students will compete to achieve the highest profit by implementing machine learning algorithms to best model the environment (e.g. users preferences, and competing agents behavior) and optimize related actions and decisions (e.g. bids on search keywords). More details on the nature of the competition are available in the Ad-Auctions competition spec.

The Trading Agents implemented will conform to a predefined modular partition. Different implementations of modules by different teams may be interchanged, implying that the performance of alternative implementations of a module may be compared. Moreover, modules from different teams may be  combined to create fully competitive agents that may take part in upcoming trading agents competitions. More details regarding the Architecture of the trading agent are available in the Architecture section.

The required background to implement the trading agents will be provided in a series of lectures:

Sponsored Internet Search - Ad Auctions

Software Infrastructure and Trading Agent Architecture

Time Series and Regression

Optimization and Machine Learning

Modeling and Particle Filtering

Reinforcement Learning