Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
SeminarAlgorithmic Game Theory
Outline for today
1. We & Our Goals2. Thesis & Peer-review3. Presentation & Feedback4. Dates & Deadlines5. You & Your Topic
Our Goals
• Ability to read and understand a scientific paper.• Extract the core points.• Scientific writing• Presentation skills.• Scientific collaborations, peer-review, feedback
• Broaden your knowledge about algorithmic game theory
Plagiarism
All citations must be marked as such and referenced to their origin. Plagiarism directly leads to failure of the seminar.
Note that plagiarism not only includes copies from the original literature, but also copies from other authors (e.g., other seminar theses) or translations of them.
Topics
• Each topic has one or several references that should be understood as starting points for your research. • you will need further papers or textbooks to fully
understand your topic. • Discuss the actual content of your seminar thesis
and seminar talk with your advisor.
Thesis
• Use Latex, detailed instructions see Website.• Essay of length 12 to 20 pages written according to
the standards of a scientific paper.• A simple summary is insufficient.• Identify the important ideas and proofs • Your readers: Your fellow students
Review Phase
• Peer review procedure similar to scientific publications• You submit your thesis (paper) at
https://easychair.org/conferences/?conf=upbagtsem18• Some (2) peers (other students) review your submission
• Read and understand the submitted paper• Criticize your paper • Make recommendations on how to improve• Recommend whether it’s worth being published
• Be honest, polite, and helpful when writing your reviews• The reviews you write will influence your final grade• The reviews you receive will not influence (but your final
version)
Presentation
• 30-35 minutes presentation• 15 minutes discussion and questions• DO not copy your thesis onto slides• Try to convey the basic ideas• Your audience: Your fellow students.• Find out how to give a good presentation for yourself• This is very different from person to person.• Some ideas e.g.
http://groups.uni-paderborn.de/matiker/Holger_Karl_matiker-talk.zip
http://groups.uni-paderborn.de/matiker/index.php@section=18
Presentation (2)• Do a test run with your presentation.• Meet in groups of four:• Group 1
• Sven Hartwig• Marcel Stienemeier• Prashanth Hariharan• Daniel Braun
• Group 2• Denis Diemert• Michael Erjemenko,• Miriam Fischer• Simon Pukrop
• Give each other feedback• Negotiate date yourselves• Tell us when and where you plan to meet (mandatory!)• We can organize a room for you if necessary
Dates
• Presentations: Day Monday July, 23rd (first day of lecture break)• Meeting in groups for test-runs July 16th or earlier• Final version theses July 16th
• Submission of theses for review: June 25th
• Review Phase June 25th to July 9th
• Start writing these s: May 8th
Short Presentations
Hello Everyone!
About me (Miriam)
I Studying Computer Science and Economics at LMU Munich
I Guest Student at UPB to get an insight into Algorithmic
Game Theory
About my topic (”Fixed Price Approximability of the Optimal Gain
From Trade”)I Mechanism Design & Approximation Algorithms
I Which mechanism approaches the optimal solution to bilateral
trade problems as close as possible? How can one improve the
approximation?
About your enthusiasm
I Mechanism & Auction Design matter!
I Benevolent dictator perspective: If you can increase social
welfare, why shouldn’t you?
I Selfish homo economicus perspective: If you sell or buy, you
want to get the most out of it!
Min-cost bipartite perfect matching with delays Ashlagi, Itai, et al, Schloss Dagstuhl-Leibniz-Zentrum für Informatik, 2017,APPROX-RANDOM 2017
(Online Algorithms)
Seminar: Algorithmic Game theory
Real world Applications • Ride sharing• Online dating websites• Job portals
Objective • Match requests• Minimize distance• Minimize delay
cost=1
cost=3
Mechanism design for one-sided markets well studied
One item, Multiple buyers, No seller role
Recent studies go more to two-sided markets
One kind of item, Multiple buyers, Multiple sellers (want to buy/sell one each)
Results mainly "what is not possible"
Paper:
Opens up two-sided scenario (think: YouTube advertising)
One kind of item ! multiple items
Users are interested in subset
Sellers sell subset
Gives good mechanisms that e�ciently approximate optimal social welfare
Defines new notions that are useful in the new scenario
Approximately E�cient Two-Sided Combinatorial Auctions
Deterministic Algorithm for Online Steiner
Tree Leasing (OSTL)
Study the Online Steiner Tree Leasing
weighted undirected graph GGoal: Lease a subset of edges connecting a given set of terminals
(a node of G) to a choosen root node r of G
Input to the problem is the sequence � of terminals arriving
sequentially in online manner
Construct a determenistic online algorithm for the OSTL problem
Competitive-ratio is subject to minimization
Result: Deterministic algorithm outperforms randomized
algorithm
Daniel Braun 07. May 2018 1
Michael Erjemenko
Shapley Facility Location Games
● SFLG are potential games● SFLG posses a PNE● Reaching efficiently a PNE via learning dynamics● Bound the Price of Anarchy
Congestion Games with Multisets of Resources
Specialisation of classical congestion gamesPlayers may need a resource more than one time∆ Congestion games with multisets for the strategiesAdds depth to the model of congestion games∆ Can be used to model systems that aren’t covered bynormal congestion gamesApplication example: System synthesis from components
1 / 1
Problem: Inconsistent behavior leading to failure of long-term goalsModel with players having present bias � 2 (0, 1]:
C = Cpresent + � · Cfuture
Describe reaching long-term goals via task graphsEstablish incentives even without knowing � exactly
1Albers S., Kraft D. (2017) The Price of Uncertainty in Present-Biased Planning.
The Price of Uncertainty in Present-Biased Planning1
Marcel Stienemeier
ɫƟNJʞȋŒɫäʞŒɻǩࢧ ɻƟɧʞƟȣƁƟɻ Œȣƌ ȴɠʉǩȝŒȋ ɻƁǞƟƌʞȋƟɻ LJȴɫ ɻƟƁʞɫǩʉˈNJŒȝƟɻࢨ ŷˈ �ƟȝɠƟࡪ ðƁǞʞȋȝŒȣࡪ üŒȝʞ˔
! ðƟƁʞɫǩʉˈ NJŒȝƟࡩ 5ƟLJƟȣƌƟɫ ɠɫȴʉƟƁʉɻ ɫƟɻȴʞɫƁƟɻ ŒNJŒǩȣɻʉ ŒʉʉŒƁȅƟɫ
! oǩNJǞȋˈ ɠɫŒƁʉǩƁŒȋࡩ! Ɵ NJࡪ ɠȋŒƁƟȝƟȣʉ ȴLJ ċð �ǩɫ �ŒɫɻǞŒȋɻࡪ ɻƁǞƟƌʞȋǩȣNJ ċð +ȴŒɻʉ cʞŒɫƌ
ɠŒʉɫȴȋɻࡪ ɻƁǞƟƌʞȋǩȣNJ LJŒɫƟ ǩȣɻɠƟƁʉǩȴȣɻ! æƟɻʞȋʉ ȴLJ ʉǞƟ ɠŒɠƟɫࡩ
ȰɛʂǦȚŒȈ ƊƝDžƝȠƊƝɥ ɛȈŒˁ ɥƝƊʗƀƝɵ ʂȰ ƀȰȚŶǦȠŒʂȰɥǦŒȈ ɛɥȰŶȈƝȚࡪ ʸǛǦƀǛƀŒȠ ŶƝ ɵȰȈʴƝƊ ƝDž˧ƀǦƝȠʂȈˁ
߾߾