Adaptive Choice of Information Sources

Slides:



Advertisements
Similar presentations
Artificial Payment Card Market: A Multi-Agent Approach Biliana Alexandrova-Kabadjova, CCFEA, EssexCCFEA Edward Tsang, CCFEA, EssexCCFEA Andreas Krause,
Advertisements

QoS-based Management of Multiple Shared Resources in Dynamic Real-Time Systems Klaus Ecker, Frank Drews School of EECS, Ohio University, Athens, OH {ecker,
TAU Agent Team: Yishay Mansour Mariano Schain Tel Aviv University TAC-AA 2010.
Reinforcement Learning
Conceptual Clustering
Making Simple Decisions
Dynamic Thread Assignment on Heterogeneous Multiprocessor Architectures Pree Thiengburanathum Advanced computer architecture Oct 24,
Resource Management §A resource can be a logical, such as a shared file, or physical, such as a CPU (a node of the distributed system). One of the functions.
Using Parallel Genetic Algorithm in a Predictive Job Scheduling
Michael R. Baye, Managerial Economics and Business Strategy, 3e. ©The McGraw-Hill Companies, Inc., 1999 Managerial Economics & Business Strategy Chapter.
Bidding Protocols for Deploying Mobile Sensors Reporter: Po-Chung Shih Computer Science and Information Engineering Department Fu-Jen Catholic University.
Randomized Sensing in Adversarial Environments Andreas Krause Joint work with Daniel Golovin and Alex Roper International Joint Conference on Artificial.
Short introduction to game theory 1. 2  Decision Theory = Probability theory + Utility Theory (deals with chance) (deals with outcomes)  Fundamental.
Static Optimization of Conjunctive Queries with Sliding Windows over Infinite Streams Presented by: Andy Mason and Sheng Zhong Ahmed M.Ayad and Jeffrey.
Planning under Uncertainty
Detecting Network Intrusions via Sampling : A Game Theoretic Approach Presented By: Matt Vidal Murali Kodialam T.V. Lakshman July 22, 2003 Bell Labs, Lucent.
Fault-tolerant Adaptive Divisible Load Scheduling Xuan Lin, Sumanth J. V. Acknowledge: a few slides of DLT are from Thomas Robertazzi ’ s presentation.
Uninformed Search Reading: Chapter 3 by today, Chapter by Wednesday, 9/12 Homework #2 will be given out on Wednesday DID YOU TURN IN YOUR SURVEY?
Exploiting Correlated Attributes in Acquisitional Query Processing Amol Deshpande University of Maryland Joint work with Carlos Sam
1 Introduction to Load Balancing: l Definition of Distributed systems. Collection of independent loosely coupled computing resources. l Load Balancing.
MEGaVis: Perceptual Decisions in the Face of Explicit Costs and Benefits Michael S. Landy Julia Trommershäuser Laurence T. Maloney Ross Goutcher Pascal.
A Multi-Agent Learning Approach to Online Distributed Resource Allocation Chongjie Zhang Victor Lesser Prashant Shenoy Computer Science Department University.
Introduction to Macroeconomics Chapter 1. An Overview of Macroeconomics.
ENGINEERING PROFESSIONALISM AND ETHICS EGN 4034 FALL 2008 CHAPTER 3-4 Organizing Principles.
Self-Organizing Agents for Grid Load Balancing Junwei Cao Fifth IEEE/ACM International Workshop on Grid Computing (GRID'04)
COGNITIVE RADIO FOR NEXT-GENERATION WIRELESS NETWORKS: AN APPROACH TO OPPORTUNISTIC CHANNEL SELECTION IN IEEE BASED WIRELESS MESH Dusit Niyato,
A Theoretical Study of Optimization Techniques Used in Registration Area Based Location Management: Models and Online Algorithms Sandeep K. S. Gupta Goran.
REINFORCEMENT LEARNING LEARNING TO PERFORM BEST ACTIONS BY REWARDS Tayfun Gürel.
IIASA Yuri Ermoliev International Institute for Applied Systems Analysis Mathematical methods for robust solutions.
1 Evaluating top-k Queries over Web-Accessible Databases Paper By: Amelie Marian, Nicolas Bruno, Luis Gravano Presented By Bhushan Chaudhari University.
Presenter: Chih-Yuan Chou GA-BASED ALGORITHMS FOR FINDING EQUILIBRIUM 1.
Balancing Exploration and Exploitation Ratio in Reinforcement Learning Ozkan Ozcan (1stLT/ TuAF)
Chapter 5 Choice Under Uncertainty. Chapter 5Slide 2 Topics to be Discussed Describing Risk Preferences Toward Risk Reducing Risk The Demand for Risky.
Conversation as Action Under Uncertainty Tim Paek Eric Horvitz.
Approximate Dynamic Programming Methods for Resource Constrained Sensor Management John W. Fisher III, Jason L. Williams and Alan S. Willsky MIT CSAIL.
Heuristic Optimization Methods Greedy algorithms, Approximation algorithms, and GRASP.
Copying distribution or use of the contents of this document is prohibited without written authorization from SafeHarbor Technology Corporation. Maximizing.
Hypothesis Testing.  Select 50% users to see headline A ◦ Titanic Sinks  Select 50% users to see headline B ◦ Ship Sinks Killing Thousands  Do people.
Optimal Testing Strategy For Product Design Wei Wang Pengbo Zhang Apr /15/20091.
Decision Making Under Uncertainty CMSC 471 – Spring 2041 Class #25– Tuesday, April 29 R&N, material from Lise Getoor, Jean-Claude Latombe, and.
1 Optimizing Decisions over the Long-term in the Presence of Uncertain Response Edward Kambour.
03/02/20061 Evaluating Top-k Queries Over Web-Accessible Databases Amelie Marian Nicolas Bruno Luis Gravano Presented By: Archana and Muhammed.
R. Brafman and M. Tennenholtz Presented by Daniel Rasmussen.
C HAPTER 2  Hypothesis Testing -Test for one means - Test for two means -Test for one and two proportions.
ECE 8443 – Pattern Recognition ECE 8527 – Introduction to Machine Learning and Pattern Recognition Objectives: Mixture Densities Maximum Likelihood Estimates.
DECISION MODELS. Decision models The types of decision models: – Decision making under certainty The future state of nature is assumed known. – Decision.
Name : Mamatha J M Seminar guide: Mr. Kemparaju. GRID COMPUTING.
Intelligent Database Systems Lab 國立雲林科技大學 National Yunlin University of Science and Technology 1 Intelligent Exploration for Genetic Algorithms Using Self-Organizing.
TEMPLE UNIVERSITY Deadline-Sensitive Mobile Data Offloading via Opportunistic Communications Guoju Gaoa, Mingjun Xiao∗a, Jie Wub, Kai Hana, Liusheng Huanga.
Authors: Jiang Xie, Ian F. Akyildiz
Introduction to Load Balancing:
Maximum Expected Utility
Efficient Join Query Evaluation in a Parallel Database System
Making complex decisions
Scheduling Jobs Across Geo-distributed Datacenters
Reinforcement Learning
Expectimax Lirong Xia. Expectimax Lirong Xia Project 2 MAX player: Pacman Question 1-3: Multiple MIN players: ghosts Extend classical minimax search.
Professor Arne Thesen, University of Wisconsin-Madison
Arne Thesen and Akachai Jantayavichit
Announcements Homework 3 due today (grace period through Friday)
Instructors: Fei Fang (This Lecture) and Dave Touretzky
CPU SCHEDULING.
   Storage Space Allocation at Marine Container Terminals Using Ant-based Control by Omor Sharif and Nathan Huynh Session 677: Innovations in intermodal.
13. Acting under Uncertainty Wolfram Burgard and Bernhard Nebel
October 6, 2011 Dr. Itamar Arel College of Engineering
Lecture 3: Environs and Algorithms
Presented By: Darlene Banta
Maximizing Speedup through Self-Tuning of Processor Allocation
Performance-Robust Parallel I/O
Games & Adversarial Search
Presentation transcript:

Adaptive Choice of Information Sources 2019-04-25 Adaptive Choice of Information Sources 96419-006 권권택

Introduction Adaptive Information Agent 정보를 수집, 가공해서 제공 사용자 취향을 학습, 저장, 가공 변화하는 환경에 유연하게 대응 We assume that agents don’t have significant control on the composition of other agents the loads on information sources they use 2019-04-25

The Goal of Adaptive Agents Decrease response time Avoid congestion in information sources Improve stability Converge to balanced, stable configuration Improve information quality Balance exploration and exploitation 2019-04-25

A Categorization of Approaches Adaptive schemes to be used by multiple information agents ; Identify lightly loaded resources. State-based solution base decision on the observed load distribution. Model-based solution consider not only the state, but also the expected behavior of other agents. 2019-04-25

Multiple adaptive agents Information Resources r-window Agents 2019-04-25

A Categorization of Approaches Adaptive schemes to be used by a stand-alone information agents Optimize the quality of information Learn about the different expertise levels or specializations of the information sources 2019-04-25

A State-based Approach The basic assumptions all loads can provide the same information response time of a source increases with its workload no explicit communication between agents 2019-04-25

A State-based Approach r-window a window through which an agent can observe the load on some resources At each time step, each agents have to decide whether continue to use the present resource or move to another resource in its r-window Use a probabilistic decision procedure 2019-04-25

2019-04-25

Results If agents are allowed access to the status of smaller number of resources, the loads on different resources are balanced in less time Convergence rate to stable configurations can be significantly enhanced if local group make their decisions sequentially 2019-04-25

Probabilistic Analysis X : number of agents who will not leave the resource in the next time step Y : number of agents who will move into the resource i in the next time step 2019-04-25

Probabilistic Analysis 2019-04-25

Probabilistic Analysis 2019-04-25

Adaptive Agents Initially use a large r-window size, but quickly reduce this size after some initial movements Improvements of adaptive scheme Skewed initial distribution : 21% Uniform initial distribution : 3% 2019-04-25

A Model-based Approach Modeling agent decision functions using Chebychev polynomials Each agent observes the load on the resource in which another agent places a job, the number of previous visits in which it did not place the job Modeling agents will use the information and find resources less likely to be selected 2019-04-25

A Model-based Approach G-agents : greedy agents P-agents : probability function agents M-agents : modeling agents 2019-04-25

2019-04-25

2019-04-25

2019-04-25

Results Addtion of M-agents reduces the standard deviation of loads across the resources Homogeneous group of M-agents do not produce effective performance Modeling scheme is able to track changes in agent behaviors 2019-04-25

Learning to Select Information Sources Different search engines are good for different kind of queries The performance of the search engine are modeled probabilistically from experience the principle of Maximum Expected Utility an agent chooses an action that yields the highest expected utility, averaged over all the possible outcomes 2019-04-25

2019-04-25

Results The MEU strategy outperforms the most-often-liked heuristic when the probability distribution for the search engines are skewed 2019-04-25

Conclusions Agents that don’t model other agents can be made to converge faster to stable distributions by introducing asynchrony. When we evaluate a mix of adaptive and static-strategy agents, everyone benefits Agents using expected utility maximizing paradigm can be used when quality information has to be returned in real-time 2019-04-25