Howard, R. A. (1966), `Information value theory', IEEE Transactions on Systems Science and Cybernetics SSC-2, 22--26.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is also a need for evaluating on beforehand whether it is worthwhile to consult an information source. Furthermore, if several sources are available there is a need to come up with a strategy for a sequence of data requests. The problem of data request has been formally treated in decision theory (Howard 1966, Lindley 1971, Shachter 1986) where utilities of possible actions are guiding the request decisions. Also utility free assessments of information sources have been studied (Ben Bassat 1978, Pearl 1988) where Shannon s measure of mutual information and variance are used as a measure of impact of ....

Howard, R.A. (1966). Information Value Theory. IEEE Transactions on Systems Science and Cybernetics , ssc2, pp 22-26.

....rapidly becomes uncertain of its position. The expected value of the resulting positions according to V MDP is poor, so the robot prefers the well lit hallway. In the medical diagnosis case, many successful diagnostic systems have been based on a one step value of information (VOI) calculation [7]. The physician considers the expected utility of choosing a therapy immediately versus the expected utility of performing one test and then choosing the therapy after the test results are known. Greedy VOI often works extremely well, both for choosing the best test to perform and for deciding ....

Howard, R. A.: Information Value Theory. IEEE Trans. Sys. Sci. and Cyber., Vol. SSC-2 (1966) 22--26

....function: we simply act optimally relative to our uncertain information about this random variable. Second, it provides a clear metric for evaluating different possible utility elicitation questions: the extent to which a question helps reach the optimal decision is simply its value of information (Howard 1966). This insight provides us with the basis for our algorithm. At every step, our algorithm computes the optimal strategy based on the current information. It then asks the elicitation question with the highest value of information, and the user s answer is incorporated into the model. This ....

Howard, R. A. 1966. Information value theory. IEEE Transactions on Systems Science and Cybernetics SSC-2:22--26.

....the epistemic payoff of an action, we see that simple economic models of costs and benefits fail because the benefits of an epistemic action depend in considerable detail on just what computations the agent is performing when undertaking an action. In classical decision analysis accounts (Howard, 1966; Raiffa, 1968) the value of a piece of information can be estimated by comparing the expected utility of an action after that information is discovered with its expected utility before. There is no need to know anything about the internal reasoning process of the agent to estimate how valuable ....

Howard, R. A. (1966). Information value theory. IEEE Transactions on Systems Science and Cybernetics, 2, 22--26.

....see, the relation between buying information and buying a lottery is more delicate than one might first guess. In Section 4 we provide some concluding remarks. 2 2 Information Value Background Early work in the area of information value as it pertains to decision analysis is attributable to Howard (1966, 1967) and Matheson (1968) Their consideration of the value of clairvoyance led to the concept of perfect information and a methodology for calculating the expected value of perfect information. General discussions of information value may be found in Raiffa (1968) Gould (1974) and Howard ....

Howard, R.A., "Information Value Theory," IEEE Transactions on Systems Science and Cybernetics, Vol. SCC-2, No. 1, p22-26, 1966.

....about Q values, they direct exploration specifically toward poorly known regions of the state space. Their approach is based on a decision theoretic approach to action selection: the agent should choose actions based on the value of the information it can expect to learn by performing them (Howard 1966). Dearden et al. propose a measure that balances the expected gains in performance from exploration in the form of improved policies with the expected cost of doing a potentially suboptimal action. This measure is computed from probability distributions over the Qvalues of actions. In ....

....models to give estimates of Q value distributions. Before we do that, we briefly review how Dearden et al. use the Q value distributions for selecting actions, as we use this method in the current work. The approach of Dearden et al. is based on the decisiontheoretic ideas of value of information(Howard 1966). The application of these ideas in this context is reminiscent of its use in tree search (Russell Wefald 1991) which can also be seen as a form of exploration. The idea is to balance the expected gains from exploration in the form of improved policies against the expected cost of doing a ....

Howard, R. A. (1966), `Information value theory', IEEE Transactions on Systems Science and Cybernetics SSC-2, 22--26.

....source of information to the current procedure (i.e. M includes L) NPVM Gamma NPVL corresponds to the net expected value of the information (NEVI) induced by the learner, 4 generalized to take time and the rate of return into account. Information value theory was originally introduced by Howard (1966) , and developed in the context of resource limited reasoning by Horvitz (1990) Here we propose applying it to the machine learning process 4 With the maximization of utility implicit in M s L s choice of class. i.e. considering the cost and value of learned information) Future Work ....

Howard, R. A. 1966. Information value theory.

....Since the 1960 s, I. J. Good [16] has emphasized the conceptual distinction between classical or type I rationality, and what he called type II rationality, or the maximization of expected utility taking into account deliberation costs. Researchers in decision analysis, especially Howard [28], have studied the problem of the value of information. Although the theory given below was developed independently, there is a strong parallel with Howard s approach. In the field of economics, Simon [46] made clear the distinction between systems that compute the rational thing to do (procedural ....

....on the passage of time spent in deliberation, since the real intrinsic utility is of course constant. This would not be the case if we used the later utility estimate for the new best move and the current utility estimate for the current best move. 9 Examining Howard s Information Value Theory [28] we find a formula which, in the above context, would amount to defining the expected net value of S j as E[ V (S j ) E[ U S:S j ( ff S j ; S j ] Gamma U S ( ff] 12) This will only be equivalent to our formula, given in equation 10, provided we have, in the current state ....

Howard, R. A. (1966) Information value theory. IEEE Transactions on Systems Science and Cybernetics, SSC-2(1), 22-26.

....is also a need for evaluating on beforehand whether it is worthwhile to consult an information source. Furthermore, if several sources are available there is a need to come up with a strategy for a sequence of data requests. The problem of data request has been formally treated in decision theory (Howard 1966, Lindley 1971, Schachter 1986) where utilities of possible actions are guiding the request decisions. Also utility free assessments of information sources have been studied (BenBassat 1978, Pearl 1988) where Shannon s measure of mutual information and variance are used as a measure of impact of ....

Howard, R.A. (1966). Information Value Theory. IEEE Transactions on Systems Science and Cybernetics , ssc-2, pp 22-26.

....rational. Procedural rationality of a grandmaster in chess comes from heuristics for selective search, and knowledge of significant patterns. Search is terminated when a satisfactory solution is found, not when the optimal solution is found (Newell Simon 1972) The theory of information value (Howard 1966) has many parallels with the work on metareasoning that we survey here. Howard proposes the idea of computing the decisiontheoretic value of an additional piece of information by simulating the decision procedure that will follow, given each possible outcome of the information request, and thereby ....

....agents should be central to AI, since perfect rationality and epistemological correctness provide neither an adequate theoretical or practical framework. Russell and Wefald (Russell Wefald 1991) present a powerful framework for metalevel rationality, using information value theory (Howard 1966), to derive conditions that define how the agent should deliberate at the base level. The external model that they promote underlies the theory of normative metareasoning. The idea is that we should take decisions at the metalevel on the basis of the value of deliberation, and that the value of ....

Howard, R. A. 1966. Information value theory. IEEE Transactions on Systems Science and Cybernetics SSC-2:22--26.

....already mentioned that Dearden and Boutilier estimate the costs and benefits of their abstractions. Nicholson and Kaelbling use a method of evaluation that accounts for the state transition probabilities but not the consequences of acting. Ideally one could use the notion of value of information [Howard, 1966] to exactly calculate the cost of adopting an abstract model, but computationally such exact calculations are infeasible. Chavez and Henrion [1994] describe a Monte Carlo method for estimating the value of information for planning abstractions. W X W X P Q , P 25 Korf [1988] ....

....problems. Pearl [1988] provides a comprehensive treatment of Bayesian networks which are a special case of influence diagrams without decision variables. Dean and Kanazawa [1989] describe how Markov processes can be compactly represented as Bayesian networks. Value of information was introduced by Howard [1966], but the basic idea has been around for much longer and is central to the field of optimal experimental design (see [Fedoras, 1972] Tatman and Shachter [1990] consider the notion of time separable value in influence diagrams. Schweitzer [1984] surveys aggregation techniques for handling large ....

30 Howard, Ronald A., 1966, Information value theory, IEEE Transactions on Systems Science and Cybernetics 2(1):22-26.

No context found.

Howard, R. A. (1966), `Information value theory', IEEE Transactions on Systems Science and Cybernetics SSC-2, 22--26.

No context found.

Howard, R. A., "Information Value Theory" IEEE Transactions on Systems Science and Cybernetics, Vol., SSC-2, No. 1, pp. 22-26, August 1966.

No context found.

Howard, R. 1966. Information Value Theory. IEEE Transactions on System Science and Cybernetics, SSC-2(1): 22-26. 14

No context found.

R. A. Howard, "Information value theory", IEEE Transactions on Systems Science and Cybernetics, 1966.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC