Core Concepts
Introducing global budget balance in bilateral trade algorithms enables the development of no-regret learning algorithms for adversarial scenarios.
Abstract
The content discusses the challenges and solutions in developing no-regret algorithms for bilateral trade under adversarial conditions. It introduces the concept of global budget balance and presents a two-phase algorithm to address the complexities of profit maximization and gain from trade. The content covers different feedback models, the importance of discretizing price spaces, and the comparison of results with prior research. Key insights include the impact of action space, partial feedback challenges, and the trade-off between profit and gain from trade.
Stats
In the full feedback model, the learner can guarantee Ë ð(â ð) regret against the best fixed prices in hindsight.
A learning algorithm guarantees a Ë ð(ð 3/4) regret upper bound with one-bit feedback.
The lower bound of Ω(ð 5/7) holds even in the two-bit feedback model.
Quotes
"The learner can guarantee Ë ð(â ð) regret against the best fixed prices in hindsight."
"A learning algorithm guarantees a Ë ð(ð 3/4) regret upper bound with one-bit feedback."
"The lower bound of Ω(ð 5/7) holds even in the two-bit feedback model."