New User Special Price Expires in

Let's log you in.

Sign in with Facebook


Don't have a StudySoup account? Create one here!


Create a StudySoup account

Be part of our community, it's free to join!

Sign up with Facebook


Create your account
By creating an account you agree to StudySoup's terms and conditions and privacy policy

Already have a StudySoup account? Login here


by: Miss Lillie Kautzer


Miss Lillie Kautzer
GPA 3.56

K. Tumer

Almost Ready


These notes were just uploaded, and will be ready to view shortly.

Purchase these notes here, or revisit this page.

Either way, we'll remind you when they're ready :)

Preview These Notes for FREE

Get a free preview of these Notes, just enter your email below.

Unlock Preview
Unlock Preview

Preview these materials now for free

Why put in your email? Get access to more of this material and other relevant free materials for your school

View Preview

About this Document

K. Tumer
Class Notes
25 ?




Popular in Course

Popular in Mechanical Engineering

This 61 page Class Notes was uploaded by Miss Lillie Kautzer on Monday October 19, 2015. The Class Notes belongs to ME 538 at Oregon State University taught by K. Tumer in Fall. Since its upload, it has received 25 views. For similar materials see /class/224519/me-538-oregon-state-university in Mechanical Engineering at Oregon State University.




Report this Material


What is Karma?


Karma is the currency of StudySoup.

You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!

Date Created: 10/19/15
03 megaquot Slat Week 3 Lecture 2 Collectives Agent Utility Functions Kagan Tumsn kagan tumevmegan ate ea may Sta Tu get utiiitiES With high facturedness and iearnabiiity start With 9 s aligned with G Gzr c is independent ofl g has cleaner signal than G Gzr c removes noise Ragan Tumsn Kagan tumevme an ateedu 03quot Greggquot 5mg Two examples for oi WOrld wlthout mequot WOrld with average mequot Kagan mm kagan tumevmegan ate edu In general agents may not be able to compute g Limited Observability Limited Computation Solutions Estimate missing information Leverage local information Approximate G orz Tradeofffactoredness vs leamability Kagan mm Kagan tumevme an ate edu Thinking in terms of Collectives39 vegem 5mg Distributed system can be designed as a collective from the ground up Routing data in a network Each router gets a new goal not shortest path routing Distributed system kept as is with a collective oating on top Data dovmload 39om a constellation of satellites A xed algorithm controls the download of ata The collective simply sets ghost traf c along the links modifying how the algorithm perceives the system Nondistributed system viewed as a collecti 39 A traditional optimization Variables viewed as age v algorithm eg simulated annealing nt man ranav man tumeimegan ate edu Example 1 Rover Coordination Rovers observe points ofinterest POls POlsvary in valuetime n Only primary observation counts Learning problem Rovers learn in single trial nonepisodic Dynamic POls appeardisappear Rovers reset at regular intervals episodic tatic POlsthe same ine i o e Dynamic POls different in each episode 7 G 7 minL 6LJle In 1 quotVm l l all dz man ranav man tumeimegan ate edu G L Global t 1 mini 6 Li 7 Lin Fully Factored LOW Learnabiljty m l l l Kagan Tumev kagan lumevmampuuns13le edu mgnn sum Global mmi 6Lj a Lin Fully Factored7 Low Learnability 3i VJ Pi g Perfectly Learnable LOW Factoredness7 00 Leamability Kaganmmen kaganlumevuveguns13lzedu Objective Functions mum39s V G Global 2 mmi Li a LN Fully Factored LOW Learnability Pi Z Z Perfectly Learnable t j 6 jquot m LOW Factoredness 00 Leamability v V Dr 1 7 7 EL 111111 6Lj7LgV 111iH 6L7Li Difference High Factoredness High Learnability Kalaquot mm kagan lumevuveguns13le all 03quot mm snag Week 3 Lecture 1 Collectives Objective Alignment in Multiagent Systems Kagan mm kagan umevme an ate edu 03 Collectives and the Design of CampIeXSystems K Turner and D Wolpert editors Springer New York 2004 ISBN 0387401652 Kagan mm Kagan mmevmegan ateedu 08quot mm snag A collective is a large multi agent system where Each agent has a private utility it is trying to optimize d A system utilityfunction measures the full system s performance Important issues How to set private utility functions How to update them team formation Can agents compute those utility functions What happens when information is missing What happens when some agents start to fail Kagan mm Kagan lumeiaregan ete edu musnine System utility Agents h Valuation of company H Employees Private Utilities H Compensation packages Design problem faced bythe board How to setmodify compensation packages private utilities of the agents to increase valuation of company system utility Salarybonus Bene ts Stock options Note Board does not tell each individual what to do They set the incentive packagesquot for employees including the CEO Regan mm Kagan lumeiare an ete edu Key Concepts for Collectives minim Factoredness Degree to which an agent s private utility is aligned with the system utility eg stock options are factored wrt company valuation Learnability Based on sensitivity of an agent s private utility to changes in its state signaltonoise eg performance bonuses increase learnability of agents utility Interesting question If you could would you want everyone s utility to be valuation of com pany Factored yes but what about learnability man runan man tumevmegan ate edu Namenclature 39EEE39ESEES39E 2 State of full system zi State of agenti zi State of full system except agent i ci Fixed vector independent of agent i zi ci Full state with fixed agent i Gz UtilityObjective forfull system giz UtilityObjective for agent i man runan man tumevmegan ate edu ureagg sats Factoredness Degree to which an agent s objective function is aligned with the system objective J L ugnz 9nz39 02 02 idz dz fz fz 12le fgn For a given 2 and discrete states 91 24 uiltltgiltzgt wit9x619 e arm A Z4 JZ Kagan Tumer kagan tumeruregunstate edu quotmassages Factoredness Degree to which an agent s objective function is aligned with the system objective a Z 5 vimlt2 eatz39mmz am 5411 GZ rJ Kagan Tumer kagan tumeruregunstate edu Oregon State Learnability Degree to which an agent s objective function is sensitive to its own actions as opposed to the background noise of other agents actions 952 yii 2 39 H042 9212 i 2ii LL97 2 2 22 My 2 2 r Mimi Qiz r A l I Z 151 il 5 High Leamabiiity LBW Loamabiiity High Leamabiiity KaganTumei kaganiumermegnnsiaie edu n D 39 Qiz a D 7 High Factoredness Low Learnabiiity Low Factoredness High Learnabiiity We h toredness igh Learnabiiity Kaganiumei kagantumermegnnstate eau Ta gel utilities with high factoredness and learnability start with megaquot 5mg g is aligned with G Gzr c is independent ofl has cleaner signal lh n G If g G dill39erenliahle then GZqci removes nelse Kagan lumen kagan lumevmegan ale edu 03quot WEBSTER39S Week 5 Lecture 1 Game theory Kagan mm kagan tumevmegan ate edu railways Assume there are two agents i and j agents are self interested they have their own preferences over the possible outcomes Assume Z is the set of those outcomes or system states i r39 Preference ordering state 2 is weakly preferred byi over 2 if 4 7 39 3 tate z is stranglypreferred byi over 2 if Preference orderin Kagan mm Kagan tumevmegan ate edu 39 lll39 l f What is utility ls utility equal to money utility money Kagan lumen kagan lumevmegan ate edu 39E l39ll l 39 The system where the agents interact Agents simultaneously choose action to perform and as a result ofthat action an outcome Z will result The actual outcome depends on the combination of action Assume each agent has just two possible actions C cooperate and D defect Kagan lumen Kagan tumevme an ate edu Given following utilities 11DD 1 uD C 1 11C D 4 11CC 4 11DD 1 uD C 391 1C D 1 11C C 4 C C gt3 39D gt D C gt DD CC zinc gt CD 2101 39 C is the rationalchoice for i Because i strongly prefers all outcomes that arise through C over all outcomes that arise through D C is the rational choice for Becausej strongly prefers al outcomes that arise through C over all outcomes that arise through D Kagan lumen kagan tumevmegan ate edu 39295395553395 Three elements Set of players Pure strategy space Payofffunctions Note A mixed strategy is a probability distribution over pure strategies Kagan mm Kagan tumevme an ate edu 39 lll39 l f 39 Characterize the game in a payoffmatrix i defect coop defect 1 4 coop 1 4 39 Agent 1 is column player 39 Agent is row player 14DD39 l uD C 1 LlCD 4 uC C 4 uDD 1 19D C 4 uCD 1 2rCCf 4 Kagan lumen kagan tumevmegan ate edu manila How does a rational agent behave in a given scenario Play Dominant strategy Nash Equilibrium strategy Pareto Optimal strategy Social welfare maximization strategy Kagan mm Kagan tumevme an ate edu llSll EEEEEES E Given any particular strategy s either C or D agent i there will be a number of possible outcomes We say s1 dominates s2 if every outcome possible by i playing s1 is preferred over every outcome possible byi playing s2 A rational agent will never play a dominated strategy So in deciding what to do we can delete dominated strategies Unfortunately there isn t always a unique undominated strategy Kagan mm Kagan tumevmegan ate edu militants Two strategies s1 and s2 are in Nash Equilibrium if Given that agent i plays s1 agentj cannot do betterthan s2 Given that agent plays s2 agent i cannot do betterthan s1 Neither agent has an incentive to deviate from a Nash Equilibrium However Not every game has a pure strategy Nash Equilibrium Some games have more than one Nash Equilibrium Kagan mm Kagan tumevme an ate edu 0 An outcome is said to be Pareto optimal or Pareto ef cient if there is no other outcome that makes one agent better off without making another agent worse of mggn 5mg 0 If an outcome is Pareto optimal then at least one agent will be reluctant to move away from it because this agent will be worse off 0 If an outcome is not Pareto optimal then there is another if I don t directly bene t from you can bene t without me suffering Kagan mm Kagan lumevmegan ale can 0 The social welfare of an outcome 2 is the sum of the utilities that each agent gets from outcome 2 manyquot 5m It captures the total amount of money in the systemquot As a solution concept may be appropriate when the whole system all agents has a single owner then overall bene t of the system is important not individuals 7 Robot coordinatlom r Traf c7 Notsomuch Kagan mm Kagan mmevmegan ale edu 0 Where preferences of agents are in opposition we have strictly competith scenarios regs 5mg 0 Zerosum encounters are those where utilities sum to a zero key feature is tnat sum is constant Setting it to zero is for normalization 0 Zero sum implies strictly competitive 0 Zero sum encounters in real life are very rare but people tend to act in many scenarios as if they were zero sum Kagan runei kagari lumevme ari ate edu 39 Two criminals are held in separate cells no communication They are toldthat mam 5m 1 One confesses and the other does not confessor is freed and the other gets 3 years 2 Both confess each gets 2 years 3 Neither confesses both get 1 year Kagan Mien kagari lumeime ari ale edu 39 Associations Payoff matrix nmgnn 5m Confess D Not Confess i defects i cooperates 2 0 J defects 2 5 5 3 J cooperates 0 3 39Top left both get punish ed for defection Bottom right reward for mutual cooperation 39Top rightbottom Ie defector gets away with it cooperator is punished Kagen runen kegen tunevmeeansete edu arsggpsme What should you do 7 The indiw39dual rational action is defect This guarantees a payoff of no worse than 2 whereas cooperating guarantees a payoff of at O Agent l defects payoffs are 2 or 5 based on agent 2 sactlons Agent l cooperates payoffs are 0 or3 based on agent 25 actions 7 So defection is the best response to all possible strategies both agents defect and get payoff 2 But intuition says this is not the best outcome 7 Surely they should both cooperate and each get payoff of 3 Kagen lumen kegen tunevmeeansate 2th There is no dominant strategy 00 is the only Nash equilibrium All outcomes except 00 are Pareto optimal CC maximizes social welfare nmggn 5mg i defects i cooperates 2 0 J defects 2 5 5 3 J cooperates 0 3 Kagan My kagan lumevmegan ale edu This apparent paradox is the fundamental problem of mult39ragent interactions In eres ted agen Real world examples nuclear arms reduction why don t I keep mine free rider systems 7 public transport The prisoner s dilemma is ubiquitous Can we recover cooperation Kagan mm Kagan lumevme an ale edu mam Sing It appears to imply that cooperation will not occur in societies of self ts 0 Conclusions that some have drawn from this analysis mggn 5mg the game theory notion of rational action is wrong somehow the dilemma is being formulated wrongly Arguments to recover cooperation We are not all machiavelli The other prisoner is my twin The shadow of the future Kagan lumen Kagan lumevmegan ale edu mam 5m 0 One solution 0 Play the game more than once If you know you will be meeting your opponent agent then the incentive to defect disappears Cooperation is the rational choice in the in nitely repeated prisoner s dilemma 0 Problems arise in n step games defect at n1 but then why not n2 etc 0 Fortunately if n is unknown this effect goes away Kagan mm Kagan lumevme an ale edu 0 Suppose you play iterated prisoner s dilemma against a range of opponents mggn 5mg 0 What strategy should you choose so as to maximize your overall payoff 0 Axelrod 1984 investigated this problem with a computer tournament for programs playing the prisoner s dilemma Kagan mm Kagan lumevmegan ale edu mam 5m ALLD Always defect the hawk strategy 0 TITFORTAT 1 On round t1 cooperate 2 On subsequent rounds t do what opponent did on round t1 0 TESTER On lst round defect If the opponent retaliated then play TIT FORTAT Otherwise intersperse cooperation amp defection 0 JOSS As TlTFORTAT except periodically defect Kagan mm Kagan lumevme an ale edu 08quot mm snag Overall winner was TlTFORTAT Why Overall Score of a strategy computed as average of performance against all other strategies TlTFORTAT was defeated by ALLD However ALLD did not do well against all opponents TlTFORTHAT won against cooperative strategies Again as in single PD being not too cooperative pays out Kagan rumev Kagan tumevmegan ate edu warrant 0 Agtltelrod suggests the following rules for succeeding in his tournament 7 Don t be envious Don t play as if it were zero sum 7 Be nice Start by cooperating and reciprocate cooperation r Retafate appropriately Always punish defection immediately but use measured force 7 don t overdo it 7 Don t hold grudges Always reciprocate cooperation immediately Kagan mm Kagan tumevme an ate edu 03U mam snag Week 4 Lecture 1 Collectives Visualizing Factoredness and Learnability Kagan Mien kagan iumevmegan aie edu 03 Kagan Tumsn Kagan mmevme an ate edu Agent State Projection Cliegnn sme KaganTumev kagan tumeruveguns13te m Agent State Projection Dmgnn sum g 2 E 2 KaganTumev kagan tumeruveguns13te m Agent State Projection quotm KaganTumev kagan tumeruveguns13te m Ag ent State Projection Imam sum Factoredness Computation KaganTumev kagan tumeruveguns13te m Analyzed Rewards 39EEEEIEESIE Pi Sum of POI values observed by agent i Gi Sum of POI values observed by all agents Di Sum of POI values observed by agent i that would have gone unobserved by other agents DiPO Di with rovers communication restricted to distance they can travel in one step 3 of space man runan man lumevmegan ate edu Dynamic Environments minim I 1 mo u 1 man runan man lumevmegan ate edu Project onto Problem Domain 08quot Projection onto Problem Domain muggy p D PO 339 I I IN I FEL I I T I 11 I i 39 L 39 g at I I 03quot warrants Recall DU Gz Gz l s actions Four modi cations BTU Gz Gz i39s actions what i can39t see TTU G Z what i can39t see GZ i39s actions what i can39t see BEU I GZ Gi39s estimate Z i39s actions EEU I Gi s estimate 2 Gi39s estimate 2 i39s actions Kagan Mien kagan tumevmegan ate edu unmade If G can be broadcast Removes not onlyi s footprints but footprints ofall agents that i cannot observe Advantage Factored Disadvantage mismatch between rst and second terms leading to low learnability Kagan mm Kagan tumevme an ate edu 03quot renewals If G cannot be broadcast Removes not onlyi s footprints but footprints of all agents that i cannot observe However agents that i cannot observe are also removed from first term Advantage learnability is better since first and second term match Disadvantage no longer factored with respect to Gz Kagan Tumeh kagan tumevmegan ate edu 39E l39ll l 39 If G can be broadcast Removes i s footprints from estimated state Advantage Factored Disadvantage mismatch between rst and second terms leading to low learnability Kagan mm Kagan tumevme an ate edu 03quot WEBSTER39S If G cannot be broadcast Removes i s footprints from estimated state In addition full state is estimated for rst term Advantage learnability is bettersince first and second term match Disadvantage no longerfactored with respect to Gz We expect however that this rst term will be much closer to G than that of TTU Kagan mm kagan tumevmegan ate edu 39E i l 39 Performance 0 8 O 01 02 03 04 05 05 O7 08 09 1 Communication Levei Kagan mm Kagan tumevmegan ate edu megaquot 5mg Performance 0 8 O 01 02 03 04 05 05 O7 08 09 1 Commumcahon Level Kagan mm kagan umevmegan a12 edu mayquot sue 0 0102030 0 050708091 Commumcatwon Leve 100 agents 1000 episodes 05 25 runs Kagan mm Kagan mmevmegan ate edu 40 Communication Lamina Tune Kagan mm kagan umevmegan a12 edu nmggn 5mg 70 Communication wruan 7mmn9mm 4m am e Leammwyne 03quot 39 l39ll le f Week 6 Lecture 2 Voting Bargaining Kagan lumen kagan lumevmegan ate edu railways Auctions Maximize profit Buyer and seller involvedaffected Voting Maximize social good Everyone involvedaffected Kagan lumen Kagan tumevme an ate edu 03quot rem Slat What happens with multiple candidates For whom should you vote What if the system is changed Plurality Borda count Instant runoff australian voting Condorcet method Kagan lumen kagan tumevmegan ate edu Voting protocols warrants Pluralit Max vote getter wins Borda count Get points for higher ranking Weighted average voting Instant runoff Australian voting Rank candidates Lowest vote getter is dropped at each round Condorcet method Rank candidates Perform pairwise runoff and tally totals Kagan mm Kagan tumevme an ate edu 03quot 39E l39ll l f Compromising rank someone higher to get himher elected Burying rank someone lower to get himher defeated Borda protocol 0 Push over rank someone higher to get someone else elected In protocol with multiple rounds Kagan lumen kagan lumevmegan ate edu warrants In a vote with 3 or more candidates one of three things must be true Vote is dictatorial There is a candidate who cannot win under any circumstance Voting is susceptible to tactical voting Since options 1 and 2 are not compatible with real life voting systems we are let with Kagan mm Kagan tumevmegan ate edu 03quot mammals Universality Complete rankings Pareto optimality If all rank X over Y outcome should rank X over Y Sovereignty Any ranking is possible Non dictatorship No one voter decides Independence of irrelevant alternatives Removing or adding nonwinner doesn t change winner 0 Kagan lumen kagan tumevmegan ate edu unmade 482AgtBgtC 40BgtCgtA o12CgtBgtA Investigate Plurality Borda Instant runoff Condorcet Kagan mm Kagan tumevme an ate edu 03quot mamas 482AgtBgtC 40BgtCgtA 12CgtBgtA Plurality vote Awins by winning 48 ofthe vote Ranking Kagan mm kagan umevmegan a12 edu rename a 482AgtBgtC 40BgtCgtA 12CgtBgtA Borda count 321 weights A 483 401121196 B 482 403 12 240 c 481 402 123 154 Ranking Kagan mm Kagan mmevmegan ate edu 03quot mamas 482AgtBgtC 40BgtCgtA 12CgtBgtA Instant runoff C is eliminated a er round 1 only 12 rst place votes Nowwe have 481A gtB I 40BgtA 12BgtA Bwins with 52vs48forA Ranking Kagan mm kagan iumevmegan ate edu rename 482AgtBgtC 40BgtCgtA 12CgtBgtA o q Condorcet AgtB4B BgtA AgtCz48 CgtA BgtC88 BgtC Ranking Kagan mm Kagan tumevmegan ate edu llSll renin seie 48 A gt B gt C 40 B gt C gt A 12 C gt B gt A Investigate Plurality A gt B gt C Borda B gt A gt C change weights Instant runoff BgtAgtC Condorcet B gt C gt A Kagan Mien kagan tumevmegan ate edu railways 48AgtBgtC 40BgtCgtA 12CgtBgtA Borda count 321 weights A483401121196 B 432 40 12 2 240 c4a14u2 12 164 Borda count 2021 weights A4B204011211012 B 432 40 2u 12 2 920 c4a14o2 12 20 3ea Borda count 543 weights A 435 40 4 12 292 B 434 40 5 12 44 c 43 s 40 4 12 5 364 Kagan Tumsn Kagan tumevme an ate edu 03quot mm snag Small market both can bene t Two people bargaining each with a preference ordering over the outcomes Example Split a given sum ofmoney First person makes offe Second person accept or rejects based on offer lf reject no one gets anything E Q m i 9 m m 02 3 Lo 2 gtlt m 3 E m Second person accepts rejects counters First person accepts rejects counters and so on Each counter reduces total by 10 Ends when one person accepts or rejects offer kagan rumev kegen tumevmegan ate edu Unique solution that satis es Invariance only preference ordering matters Anonymity agents are treated the same way no discrimination rang12559 Pareto efficiency can t improve one without make another Independence ofirrelevant alternatives removing outcomes doesn t change result Maximize Kagan mm kegen tumevme an ate edu nmgnn 5mg Week 2 Lecture 2 MultiAgent Systems Kagan rumev Kagan lumevmegan ale edu Distributed Computing Parallelization synchronization Information is distributed control generally is not mam 5m Distributed Al Problem solving communication nformation and control distributed Distributed Problem Solving Task Decomposition Information and control distributed Distributed Control Local solutions synchronization Control is distributed information may not be Multiagent Systems Coordination interaction Simple behavior complex inter ctiuns No guarantees about other agents or system interactions Limited synchronization communication decomposition Kagan mgr Kagan lumevme an ale edu mgnn State Design autonomous agents that Interact with one another Have limited observation about the environment Have limited communication Have only local control capabilities Interact in asynchronous manner Analyze autonomous agents possibly preexisting that Have all the properties listed above Keyissues No centralized control No synchonization kaeen runen kagan lumelmegon ate edu rem 5m Hierarchy task and resource allocation propagates down 39om above W a aks down Who solves full problem Tep agent7 Centralized solution Market Bids for tasks and resources lnherentty distributed Rubustt 2 0S M System interactluns7 l vlslbie hand7 Specialization tasks decomposed based on resources capabilities lnherentty distributed Who solves the full problem Systeminteractiuns kaeen lumen kagan tumelore on ate edu areggn 5m Communication Who talks to whom Tm Information from other agents reliable Trading agents Team Ufrnbuts Modeling Do agents know what other agents will do Team Formation Cooperation shared bene ts Coalitions Cooperation selfinterest Kagan mm kagan lumevmegan ate edu mungquot 5m The system performance can be measured by Utility Jnctions Economicsgametheory Reward Jnctions Reinforcement Learning Objective functions Optlmiza Ion Goals Planning Evaluation functions Evolutionary algorithms Kagan mm Kagan tumevme an ate edu 39 i39 li f Benefl Selfinterested agents Agents maximize their own performance IS Simpietu pulse rubiem Lucaiactiuns iucaiperfur Emeasure L v age Ecunumics sumaismenceinsights Iss SI Whatnappenstuginbaibehaviur v Tragedy uftne an n 7 Dupiicatiunuftasks Potential solul Mafket b ions ased malnuds Negatiatiuns Cuiiemves Kagan mm kagan iumevmegan a12 edu 08quot 99555 Week 6 Lecture 1 Negotiations Auctions Kagan lumen Kagan tunievmegan ate edu quot995553395 How do agents reach agreements when they are self interested In an extreme case zero sum encounter no agreement is possible but in most scenarios there is potential for mutually bene cial agreement rest on matters of common inte The capabilities of negotiation and argumentation are central to the ability of an agent to reach such agreements Issues in Negotiation Protocols Mechanisms Strategies Kagan mm Kagan tunievme an ate edu llSll EEEEEES E Negotiation is governed by a particular mechanism or protocol The mechanism defines the rules of encounter between agents Mechanism design is designing mechanisms so that they have certain desirable properties Given a particular protocol how can a particular strategy be designed that individual agents can use Kagan lumen Kagan tumevmegan ate edu quot995553395 Desirable properties of mechanisms Convergenceguaranteed success Maximizing social welfare Pareto efficiency Distribution Kagan mm Kagan tumevmegan ate edu EEEEEES E An auction takes place between an agent known as the auctioneer and a collection of agents known as the bidders The goal of the auction is for the auctioneer to allocate the good to one of the bidders In most settings the auctioneer desires to maximize the price bidders desire to minimize price Auctions differ by Winner determination Secrecy of bids Auction procedure Kagan lumen Kagan tunievmegan ate edu arengn lati Goods can have private value publiccommon value correlated value Winner determination may be first price second price Bids may be open cry sealed bid Bidding may be a ndin descending Kagan mm Kagan tumevme an ate edu 03quot EEEEEES E Good s valuation 39 Good has public common value Good has the same value for all bidders eg iphone 0 Good has private value Good has different value for each agent eg Grandpa s chair 0 Good has correlated value Value of good depends on own private value and private value for other agents eg Buy with intention to sell it later Winner determination 0 First price Highest bid wins Winner pays his bid Second price Highest bid wins Winner pays secondhighest bid Kagan lumen Kagan lumevmegan ate edu Secrecy of bids militant 39 Open cry All agents know all agents bids 39 Sealed bid No agent knows other agents bids Auction procedure 0 One shot Only one bidding round 0Ascending Auctioneer begins at minimum price bidders increase bids 39 Descending Auctioneer begins at price over value of good and lowers the price at each round Kagan mm Kagan tumevme an ate edu English Auctions mm 5115 Procedure auctioneer starts by suggesting a reservation price Agents bid higher than previous bid on the good u 4 n 4 to u Most commonly known type of auction first price o ascending Dominant strategy to successively bid a small amount more than the current highest bid until it reaches agent s valuation then withdraw curse Why didnY anyone bid higher Did I obverbid ills spum39ous bidders Who raise the price man ranan man lumelmegan ale edu Dutch Auctions Procedure auctioneer starts by offering good at arti cially high value auctioneer lowers offer price until some agent makes a bid equal to the current offer rice the good is then allocated to the agent that made the offer Dutch auctions are examples of First price opencry Descending No dominant strategy seu oquotdominant strategy Assuming that others bid their true valuation Bid less than true valuation L IM 39 rimmumw bId money man ranan man lumelmegan ale edu quot3 winner First Price Sealed Bid Auctions Dragon sma Procedure There is a single round Agents get one chance at putting a bid w fnr the mount ofthe winner39s bid First price sealed bid auctions are examples of First rice ea One shot No dominant strategy pseudo dominant strategy Assuming that others bid theirtrue valuation Bid less than true valuation Bid of Agent is winner I i 39 Iaral 39 fw adWmIIBi bld money man ranan man tumevmegan aie edu Vickrey Auction WEBER Procedure There is a single round Agents get one chance at putting a bid k39 k mu m the price ofthe second highest bid First price sealed bid auctions are examples of One shot Dominant strate Bid your true valuation Why7 it yuu bid less than yuurtrue valuatiun yuu unly Bid of I1 I9 I3 I6 Agent Winner pm paid L I I I bid money man ranan man tumevmegan aie edu Auctions 39EEERESS E Auctioneerwants to maximize his revenue Which auction mechanism should he choose If bidders are risk neutral All four auctions will give same revenue under certain assumptions If bidders are riskaverse bidders would accept to pay slightly above own true valuation Dutch and rstpricesealedbid give max revenue Bidders can insure themselvesquot by bidding slightly more than true valuation If auctioneer is risk averse Vickrey amp English are best Kagan mm Kagan iumevme an ate edu Phone Call Competition Example minim Customer wishes to place longdistance call Carriers simultaneously bid sending proposed prices Phone automatically chooses the carrier dynamically ll 020 M3 Kagan mm kagan iumevmegan ate edu Best Bid Wins mamas Phone chooses carrier with lowest bid Carrier gets amount that it bid Kagan mm Kagan tumevmegan ate edu Attributes of the Mechanism 39EME39EEEES E Distributed Symmetric Carrlers have an j Stabe incentive to J Simple invest effort in strategic J EffClem behavior Maybe I can bid as high as 021 Kagan mm kagan tumevmegan ate edu Best Bid Wins Gets Second Price Iickrey Auction rewrites Phone chooses carrier with lowest bid Carrier gets amount of secondbest price Kagan mm Kagan tumevmegan ate edu Attributes of the Vickrey Mechanism minim i D Str bm d Carriers have no i Symmetric incentive to 1 Stable invest effort in 1 Simple strategic 1 Ef cient behavior ATampT 000 020 I have no e reason to 018 023 overbid Kagan mm kagan tumermegan ate edu Lies and Collusion llSll renewI5 The various auction protocols are susceptible to lying on the part of the auctioneer and collusion among bidders to varying degrees A four auctions English Dutch FirstPrice Sealed Bid Vickrey can be manipulated by bidder collusion A dishonest auctioneer can exploit the Vickrey auction by lying about the 2quotdhighest bid Shils can be introduced to inflate bidding prices in English auctions man runav man tumevmegan ete edu Negotiation quot995533395 Auctions are only concerned with the allocation of goods richer techniques for reaching agreements are requ39red Negotiation is the process of reaching agreements on matters of common intere Any negotiation setting will have four components A negotiation set possible proposals that agents can make A protocol Strategies one for each agent which are private A rule that determines when a deal has been struck and what the agreement deal is Negotiation usually proceeds in a series of rounds with every agent making a proposal at every round man runar man tumevme an ete edu reunites Week 1 Lecture 2 Agents Intelligence Environments Kagan mm kagan tumevmegan ate edu Two de nitions An agent is a computer system that is capable ofindependent autonomous acti n Autonomous gure out what needs to be done How to select ac io ns How to evaluate actions An agent is a computer system that senses reasons about and acts in an environment nsing Reasoning Acting Environment Kagan mm Kagan tumevme an ate edu remind Complexity oftasks that we automate has grown A lot of what we take for granted today would have been viewed as Al ten years ago Autopilot ofa 747 Dee BI ue Internet searches Seems Intelligence is something offin the distance but in reality it is in many everyday products My de nition An agent that senses the world reasons ab out the world acts within that world and learns from its interaction with the world is an intelligence agent Kagan mm kegen tumevmegan ate edu manila De nition A multiagent is a system that consists ofmultiple agents that interact with one another and the environment Multiple Agents Interact Environment Kagan mm Kegen tumevme an ate edu 03 megaquot 5mg Isn t itjust Distributed systems AI Gametheory Economicsmechanism design Social science Biologicalecological modeling Kagan mm kagan tumevmegan ale edu may in Accessible vs inaccessible Deterministic vs nondeterministic Episodic vs nonepisodic Static vs dynamic Discrete vs continuous Kagan mm Kagan tumevmegan ale edu ragga Slat Accessible vs inaccessible Accessible Agents can obtain complete accurate up to date information about the environment Most environments ofinterest are inaccessible For example anything operating in the real world Kagan rumev kagan lumevmegan ale edu 03 mam 5m Deterministic vs nondeterministic Deterministic an action taken by an agent has a predictable ce There is no uncertainty about the outcome of an ns action Most environments ofinterest are nondeterministic For example most things operating in the real world Kagan mm Kagan tumevmegan ale edu 39EEEEEES E Episodic vs nonepisodic Episodic The agent acts for a xed number oftime steps and then the world is reset There is no link between the episodes Nonepisodic The agents operates continuously in an environment Some real world problems are episodic games Other real world problems are nonepisodic exploring a terrain Some problems can be viewed as either 7 R b u at trying tn find a target it a r Kagan mm Kagan tumevmegan ale edu areggpsm Static vs dynamic Static The environment stays the same except for changes caused by the agents actions A robot aims to detect xed goals in an arena Dynamic The environment in which the agent operates changes Goals that robot needs to detect appeardisappearlmove Kagan mm Kagan tumevmegan ale edu 03 nmggn 5mg Discrete vs continuous Discrete There is xed nite number of actions A game of chess is discrete discrete does not mean easyquot Continuous There is an in nite number of actions 7 Directiun angie 7 Speed Kagan mm kegen tumevmegan ate edu


Buy Material

Are you sure you want to buy this material for

25 Karma

Buy Material

BOOM! Enjoy Your Free Notes!

We've added these Notes to your profile, click here to view them now.


You're already Subscribed!

Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'

Why people love StudySoup

Steve Martinelli UC Los Angeles

"There's no way I would have passed my Organic Chemistry class this semester without the notes and study guides I got from StudySoup."

Jennifer McGill UCSF Med School

"Selling my MCAT study guides and notes has been a great source of side revenue while I'm in school. Some months I'm making over $500! Plus, it makes me happy knowing that I'm helping future med students with their MCAT."

Bentley McCaw University of Florida

"I was shooting for a perfect 4.0 GPA this semester. Having StudySoup as a study aid was critical to helping me achieve my goal...and I nailed it!"


"Their 'Elite Notetakers' are making over $1,200/month in sales by creating high quality content that helps their classmates in a time of need."

Become an Elite Notetaker and start selling your notes online!

Refund Policy


All subscriptions to StudySoup are paid in full at the time of subscribing. To change your credit card information or to cancel your subscription, go to "Edit Settings". All credit card information will be available there. If you should decide to cancel your subscription, it will continue to be valid until the next payment period, as all payments for the current period were made in advance. For special circumstances, please email


StudySoup has more than 1 million course-specific study resources to help students study smarter. If you’re having trouble finding what you’re looking for, our customer support team can help you find what you need! Feel free to contact them here:

Recurring Subscriptions: If you have canceled your recurring subscription on the day of renewal and have not downloaded any documents, you may request a refund by submitting an email to

Satisfaction Guarantee: If you’re not satisfied with your subscription, you can contact us for further help. Contact must be made within 3 business days of your subscription purchase and your refund request will be subject for review.

Please Note: Refunds can never be provided more than 30 days after the initial purchase date regardless of your activity on the site.