diff --git a/year4/semester2/CT421/notes/CT421.pdf b/year4/semester2/CT421/notes/CT421.pdf index c0c2dce0..8a1dc3a4 100644 Binary files a/year4/semester2/CT421/notes/CT421.pdf and b/year4/semester2/CT421/notes/CT421.pdf differ diff --git a/year4/semester2/CT421/notes/CT421.tex b/year4/semester2/CT421/notes/CT421.tex index 22811eee..f5758336 100644 --- a/year4/semester2/CT421/notes/CT421.tex +++ b/year4/semester2/CT421/notes/CT421.tex @@ -736,5 +736,148 @@ We sill have some problems though: \item In auctions, agents agree on a price; can we deal with more dimensions of negotiation? \end{itemize} +\section{Automated Negotiation} +\textbf{Negotiation} is a means for a group to arrive at an agreement. +It is a process of joint decision-making where parties with different preferences seek to reach a mutually acceptable solution. +It is a fundamental mechanism in multi-agent systems \& human society. +Negotiation research deals with three topics: +\begin{itemize} + \item Negotiation protocols; + \item Negotiation objects; \& + \item Agents' decision-making models. +\end{itemize} + +\subsection{Negotiation Protocols} +\textbf{Negotiation protocols} are a set of rules that govern the interaction: +\begin{itemize} + \item Includes the permitted type of participants; + \item Negotiation states; + \item Events that change states; + \item Actions of participants; + \item Rules for agreement formation; \& + \item Termination conditions. +\end{itemize} + +\subsection{Negotiation Objects} +\textbf{Negotiation objects} consist of a range of issues over which agreement must be reached. +Related issues include the type of operations on agreements and altering the structure of the negotiation. + +\subsection{Agents' Decision-Making Models} +The \textbf{agents' decision-making models} are influenced by protocol, the nature of the negotiation objects, \& the range of operations. +The relative importance of the components varies depending on the domain. + +\subsection{Domain Variation in Negotiation} +In some domains, the negotiation protocol is the dominant concern. +For example, in some auction settings, the best strategy for an agent is to bid to their true evaluation --- hence no strategic analysis is really required. +In other domains, the converse is true; +given the wide range of possibilities, there is no best technique for automated negotiation. + +\subsection{Negotiation as Distributed Search} +Negotiation can be viewed as a \textbf{distributed search} through a space of potential agreements; +the dimensionality \& topology of this space is determined by the structure of the negotiation object. +One could consider each attribute of the negotiation object to have a separate dimension associated with it. +As dimensions are added (or removed), the number of points of agreement may increase (or decrease). +Similarly, if an agent changes one of the values, it is moving from one point in the agreement space to another. +\\\\ +In a negotiation, participants are the active components that determine the direction of the search. +Initially, each agent will have a portion of the space in which it will be willing to make an agreement. +Also, agents will have some means to rate the points in the space. +Negotiation involves the agents suggesting points or spaces. + +\subsection{Minimal Negotiation Capabilities} +The \textbf{minimal negotiation capabilities} are: +\begin{itemize} + \item To propose some point of space as being acceptable; \& + \item To respond to such a proposal by indicating whether or not it is acceptable. +\end{itemize} + +A simple setting is a Dutch auction: +\begin{itemize} + \item One agent (the auctioneer) calls out prices. + \item If there is no signal of acceptance by an agent, then the auctioneer makes a new offer which it believes will be more acceptable. + \item The process repeats. +\end{itemize} + +If agents can only accept or reject offers, the negotiation will be very time-consuming \& inefficient. +The proposer is effectively picking points in the agreement space based on what it perceives \& hopes to stumble upon correct point. +For negotiation to be more efficient, the recipient needs to offer feedback. + +\subsection{Feedback in Negotiation} +\textbf{Feedback} can be a critique, or a counter-proposal. +A \textbf{critique} provides two forms of feedback: +it suggests constraints on issues, and indicates acceptation or rejection of particular negotiation issues. +The more information placed in the critique, the easier it is for the original agent to determine the boundaries of the agreement space. +\\\\ +A \textbf{counter-proposal} is a proposal that is more favourable to the sender, made in response to a previous proposal. +It can suggest amendments or additions, provides implicit information about preferences, and can significantly speed up the negotiation process. + +\subsection{Limitations of Simple Proposals} +Proposals, critiques, \& counter-proposals are mere statements of what the agents want; hence, the scope is confined. +Agents can't justify their negotiation stance or persuade one another to change the negotiation stance. +This leads to the idea of \textbf{argumentation-based negotiation}: allow agents to offer more information than available proposals, critiques, \& counter-proposals. + +\subsection{Approaches to Negotiation} +Approaches to reasoning in a negotiation setting can be loosely categorised as game-theoretic, heuristic, or argumentation-based. + +\subsubsection{Game-Theoretic Approach} +The \textbf{game-theoretic approach} can be applied in two manners: +\begin{itemize} + \item Designing appropriate protocols that will govern agent's interactions; \& + \item Design of a particular agent's strategy. +\end{itemize} + +Properties of the game-theoretic approach include: +\begin{itemize} + \item We usually assume that a rational agent will choose the best strategy; + \item Finding the best strategy can be computationally intractable; \& + \item Disadvantages include that it may be difficult to characterise agent's preferences with respect to all possible outcomes. +\end{itemize} + +\subsubsection{Heuristic Approach} +The \textbf{heuristic approach} seeks to search the negotiation space in a non-exhaustive fashion. +It produces good solutions rather than optimal solutions. +Disadvantages of the heuristic approach include: +\begin{itemize} + \item Sub-optimality: it adopts an approximate notion of rationality and does not fully examine the negotiation space; + \item Models need extensive analysis true simulation; \& + \item It is usually difficult to predict behaviour. +\end{itemize} + +\subsubsection{Argumentation-Based Approach} +In the \textbf{argumentation-based approach}, agents aim to persuade or change the opponent's ratings over the agreement space. +Additional information is provided in addition to proposals, etc., usually taking the form of: +\begin{itemize} + \item \textbf{Threats:} ``if you don't accept, I'll have to...''; + \item \textbf{Rewards:} ``if you accept this offer, in the future I'll...''; \& + \item \textbf{Appeals:} ``this is standard practice in our industry...''. +\end{itemize} + +\subsection{Extended Topics in Negotiation} +\subsubsection{Learning in Negotiation} +Agents can adapt strategies based on past interactions. +Types of learning: +\begin{itemize} + \item Learning opponent's preferences; + \item Learning effective negotiation strategies; \& + \item Learning from past negotiation outcomes. +\end{itemize} + +\subsubsection{Trust \& Reputation} +\textbf{Trust \& reputation} are critical when agreements must be enforced overtime. +\textbf{Trust} models help agents to decide with whom to negotiate. +\textbf{Reputation} systems aggregate experiences across multiple agents. +There are mechanisms for: +\begin{itemize} + \item Preventing reneging on agreements; + \item Handling deception in negotiation; \& + \item Building long-term relationships. +\end{itemize} + + + + + + + \end{document}