LYU 0004 Mobile Agent’s Community Group Member: Cheng Tsz Hei Ho Man Lam
Outline of the Presentation Project’s system architecture Introduction multi-sellers & multi-buyers scenario The shortcoming of traditional approach The advantage of our system over the old one
Outline of the Presentation Algorithm used to handle the communication Difficulties Future plans
Mobile’s Agent The mobile agents can act on behalf of the user in the computer network. Mobile agents are programs that can be dispatched from one computer and transported to a remote computer for execution.
Mobile Agent’s Community Group of agents with different purposes Several network computers support agent’s platform Easy to be accessed by web client Form a virtual community
Model of System Architecture Front-end
Model of System Architecture (II) Front-end
Model of System Architecture (III) Developer’s View Internet
Model of System Architecture (IV) Developer’s View Internet
Model of System Architecture (V) Developer’s View
Model of System Architecture (VI) Front-end & Developer’s View
Model of System Architecture (VII) Inside Workplace
Multi-sellers & Multi-buyers Scenarios In this scenario, the buyers or the sellers can assign their trade strategies by using graphical user interfaces in the web site. In this scenario, the buyers or the sellers can assign their trade strategies by using graphical user interfaces in the web site. One of workplaces connecting to the web then delegates mobile agents to autonomously perform the bargain behavior for the client. One of workplaces connecting to the web then delegates mobile agents to autonomously perform the bargain behavior for the client.
Existing Approach Electronic market Electronic market Fix web server Fix web server Applet or CGI technology Applet or CGI technology Fully controlled by users Fully controlled by users
Shortcoming of Traditional Approach Central access point Central access point Deficiency of interaction Deficiency of interaction Transaction localization Transaction localization
Central Access Point One web server One web server Low response time Low response time Traffic jam for local region of network Traffic jam for local region of network
Deficiency of Interaction Seller waits for buyer & vice versa Seller waits for buyer & vice versa Time consuming Time consuming Easy to miss time slot Easy to miss time slot
Transaction Localization Location boundary Location boundary Limit the potential clients Limit the potential clients Hard to promote globally Hard to promote globally
Advantage of Our System Location transparency Location transparency Failure transparency Failure transparency Scaling transparency Scaling transparency Fast response Fast response
Location Transparency Hide the real location of marketplace Hide the real location of marketplace Agents will locate the paths of possible marketplaces Agents will locate the paths of possible marketplaces
Failure Transparency Redundancy Redundancy Agents move from one marketplace to another one Agents move from one marketplace to another one No transactions are suspended and discarded No transactions are suspended and discarded
Scaling Transparency Build up list of address of workplaces Build up list of address of workplaces Allow to join or leave at any time Allow to join or leave at any time Expands infinity Expands infinity
Fast Response Biding representative to their clients in the bargaining sites Biding representative to their clients in the bargaining sites Immediate response according to client’s preference Immediate response according to client’s preference Maximize profit for both buyers and sellers Maximize profit for both buyers and sellers
Concept of Algorithm Zero-Intelligence-Plus (ZIP) Traders Zero-Intelligence-Plus (ZIP) Traders Dave Ciff, Hewlett Packard Laboratories, Bristol, England, 1997 Dave Ciff, Hewlett Packard Laboratories, Bristol, England, 1997 Act as double auction market Act as double auction market Behave as human market Behave as human market Obeys the theory of supply and demand Obeys the theory of supply and demand
Concept of Algorithm (II) Limit price is private Limit price is private Shout-prices observed in the market Shout-prices observed in the market Each agent adjust its margins up or down Each agent adjust its margins up or down Accept or ignore Accept or ignore
Algorithm for Trading Seller Behaviors. if (the last shout was accepted at price q). then any seller s i for which p i <= q should raise its profit margin. any seller s i for which p i <= q should raise its profit margin. if(the last shout was a bid). if(the last shout was a bid).then. any active sellers s i for which p i >= q should lower its margin. else. if the(last shout was an offer). then any active seller s i for which p i >= q should lower its margin. any active seller s i for which p i >= q should lower its margin.. where q is the shout price of the last shout. p i is shout price of trader i. p i is shout price of trader i.
Algorithm for Trading (II) Buyer Behaviors if (the last shout was accepted at price q) if (the last shout was accepted at price q)then any buyer b i for which p i >= q should raise its profit margin any buyer b i for which p i >= q should raise its profit margin if(the last shout was a offer) if(the last shout was a offer)then any active buyers b i for which p i <= q should lower its margin any active buyers b i for which p i <= q should lower its marginelse if the(last shout was an offer) if the(last shout was an offer) then then any active buyer b i for which p i <= q should lower its margin any active buyer b i for which p i <= q should lower its margin where q is the shout price of the last shout. p i is shout price of trader i p i is shout price of trader i
Algorithm for Trading (III) How to shout price P i (t) ? At time t,P i (t) =λ i,j (1+μ i (t)) where λ i,j is limit price, μ i (t) profit-margin For seller, profit margin constraint between 0 <= μ i (t) < ∞ For buyer, profit margin constraint between -1<= μ i (t) < 0
Algorithm for Trading (III) How to calculate the profit-marginμ i (t+1)? Using Widrow-Hoff “delta rule”: μ i (t+1) = (P i (t) + T i (t)) /λ i,j – 1 where T i (t) momentum-based update T i (0) = 0 for all i
Difficulties Problem domain Problem domain Configuration Configuration Implementation of grasshopper Implementation of grasshopper Weakness of java Serlvet Weakness of java Serlvet
Future Plan Real-time interaction Real-time interaction Learning technique for agents Learning technique for agents Higher-order adaptation mechanisms Higher-order adaptation mechanisms Game-theory analysis Game-theory analysis Support more scenarios in mobile agent’s paradigm Support more scenarios in mobile agent’s paradigm Security issue Security issue Professional design of web site Professional design of web site
The End