Algorithmic game theory (NDMI098) - lecture


Time of the lecture: Thursday 2:00pm, in the room S9.

Instructor: Martin Balko. E-mail: balko (AT) kam.mff.cuni.cz

Tutorials.


Information:
up
  • 2/2, C+Ex, 5 E-Credits
  • Annotation: An introduction to algorithmic game theory, a relatively new field whose objective is to study formal models of strategic environments and to design effective algorithms for them. This introductory course covers basic concepts and methods that are illustrated with several practical applications. To successfully pass the course, it is recommended to know basics from complexity theory.
  • Throughout the winter term, I will keep posting lecture notes about topics covered so far.
  • Literature:
    • Noam Nisan, Tim Roughgarden, Éva Tardos, and Vijay V. Vazirani, editors. Algorithmic game theory. Cambridge University Press, Cambridge, 2007.
    • Tim Roughgarden. Twenty lectures on algorithmic game theory. Cambridge University Press, Cambridge, 2016.
    • Kevin Leyton-Brown and Yoav Shoham. Essentials of game theory, volume 3 of Synthesis Lectures on Artificial Intelligence and Machine Learning. Morgan & Claypool Publishers, Williston, VT, 2008.
    • Jiøí Matoušek and Bernd Gärtner. Understanding and Using Linear Programming. Springer-Verlag New York, Inc., 2006.
  • Lecture notes: see the current run of the course.
    • The lecture notes are still under construction. If you notice any mistake or place for improvement, please, let me know by e-mail.

Lectures:
up
  • First lecture (7.10.2021):
    • Introduction, information about the credit and the exam,
    • normal-form games, Nash equilibria (pure and mixed), examples of normal-form games,
    • Nash's theorem and we started its proof using Brouwer's fixed point theorem,
    • presentation [PDF].
  • Second lecture (14.10.2021):
    • Proof of Nash's theorem,
    • Pareto optimality,
    • the Minimax theorem and preparations for its proof using the duality in linear programming,
    • presentation [PDF].
  • Third lecture (21.10.2021):
    • Proof of the Minimax theorem based on the duality of linear programming,
    • best response condition,
    • brute-force algorithm to find all Nash equilibria,
    • presentation [PDF].
  • Fourth lecture (28.10.2021):
    • The lecture is cancelled (Independent Czechoslovak State Day).
  • Fifth lecture (4.11.2021):
    • Best response polyhedra,
    • best response polytopes,
    • Lemke–Howson algorithm and a proof of its correctness,
    • presentation [PDF].
  • Sixth lecture (11.11.2021):
    • Complexity classes FNP and PPAD,
    • NASH being FNP-complete implies NP = coNP (without proof),
    • the END-OF-THE-LINE problem,
    • NASH is PPAD-complete (without proof),
    • epsilon-Nash equilibria,
    • quasi-polynomial time algorithm for finding epsilon-Nash equilibria,
    • presentation [PDF].
  • Seventh lecture (18.11.2021):
    • Correlated equilibria and their properties,
    • linear program for finding correlated equilibria.
    • Regret minimization, introduction of the formal model, external regret,
    • large comparison classes cannot yield good bound on external regret,
    • greedy algorithm and its cumulative loss,
    • presentation [PDF].
  • Eighth lecture (25.11.2021):
    • Randomized greedy algorithm and its cumulative loss,
    • polynomial weights algorithm and its cumulative loss,
    • no-regret dynamics,
    • proof of the Minimax theorem using regret minimization,
    • presentation [PDF].
  • Ninth lecture (2.12.2021):
    • Coarse correlated equilibria,
    • convergence to coarse correlated equilibria with no-regret dynamics,
    • internal regret and swap regret,
    • black-box reduction producing good bounds on swap regret using algorithms with good bounds on external regret,
    • no-swap-regret dynamics and its convergence to correlated equilibria,
    • presentation [PDF].
  • Tenth (double) lecture (9.12.2021):
    • Mechanism design,
    • single-item auctions,
    • first-price auction, Vickrey auction,
    • DSIC property, Vickrey auction is awesome,
    • single-item environments, sponsored-search auctions,
    • statement of Myerson's lemma,
    • applications of Myerson's lemma (single-item auctions and sponsored-search auctions),
    • sketch of the proof of Myerson's lemma,
    • Knapsack auctions,
    • almost optimal greedy algorithm for Knapsack auctions (with a sketch of the proof of correctness),
    • presentation [PDF].
  • Eleventh lecture (16.12.2021):
    • Revenue maximization,
    • Bayesian model and expected revenue,
    • maximizing expected revenue = maximizing expected virtual social surplus,
    • maximizing revenue in single-item auctions and further extensions,
    • presentation [PDF].
  • Twelfth lecture (6.2.2022, recorded lecture from last year):
    • Bulow–Klemperer Theorem,
    • Revelation principle,
    • multi-parameter environments and their examples,
    • VCG mechanism,
    • presentation [PDF].

Valid XHTML 1.0 Transitional