Hiroshige Inazumi

A class of distortionless codes designed by Bayes decision theory

The problem of distortionless encoding when the parameters of the probabilistic model of a source are unknown is considered from a statistical decision theory point of view. A class of predictive and nonpredictive codes is proposed that are optimal within this framework. Specifically, it is shown that the codeword length of the proposed predictive code coincides with that of the proposed nonpredictive code for any source sequence. A bound for the redundancy for universal coding is given in terms of the supremum of the Bayes risk. If this supremum exists, then there exists a minimax code whose mean code length approaches it in the proposed class of codes, and the minimax code is given by the Bayes solution relative to the prior distribution of the source parameters that maximizes the Bayes risk. >

A new scheme of case-based decision support systems by using DEA and GA techniques

We propose a new scheme for case-based decision support systems (DSS) using data envelopment analysis (DEA) and genetic algorithms (GA). The application field for our scheme is cases of multiple input-output activity in which the efficiency of the outputs is evaluated. The case-based DSS offer activity-policy which reflects any level, both of the efficiency and features in the activity, referring to many past cases in the same area. Our scheme is based on two procedures, an analysis procedure and an estimation procedure. In the analysis procedure, all the cases are recursively evaluated by solving the modified model of DEA, i.e., the generalized BCC model. The remaining cases, except for cases belonging to the efficiency frontier, are also evaluated by the generalized BCC model, and the processing is repeated. After the analysis procedure, it is possible to classify the cases into multiple hierarchies by level of efficiency, and also into the groups with common features between inputs and output, which cover multiple hierarchies. In the estimation procedure, according to past cases, features, frontier levels, and required conditions, any solution set of future activity plans is searched by a GA with a fitness function using these factors. After the estimation procedure, the user controls the variety of required conditions about past cases of activity, and finally chooses the future plan.

An inductive inference procedure to minimize prediction error

Considering inductive inference and deductive inference as not individual processes but a serial process of information processing, the serial inference procedure is studied for two purposes: the compression of observed facts and the prediction of new well-formed formulas. A serial inference process scheme that uses the correspondence to source coding and prediction problems is proposed. The optimal inference procedures for the two purposes are shown in the proposed scheme. >

System evaluation of construction methods for multi-class problems using binary classifiers

Construction methods for multi-valued classification (multi-class) systems using binary classifiers are discussed and evaluated by a trade-off model for system evaluation based on rate-distortion theory. Suppose the multi-class systems consisted of \(M (\ge 3)\) categories and \(N (\ge M-1)\) binary classifiers, then they can be represented by a matrix W, where the matrix W is given by a table of M code words with length N, called a code word table. For a document classification task, the relationship between the probability of classification error \(P_e\) and the number of binary classifiers N for given M is investigated, and we show that our constructed systems satisfy desirable properties such as “Flexible”, and “Elastic”. In particular, modified Reed Muller codes perform well: they are shown to be “Effective elastic”. As a second application we consider a hand-written character recognition task, and we show that the desirable properties are also satisfied.

A new scheme for verifying rule-based systems using Petri nets

In this paper, we propose a new scheme for verifying rule-based systems using Petri nets. Especially, considering the property of Transition-invariant (T-invariant), the simple method of detecting structural errors in rule-based systems (RBS) is clarified. That is to say, some reasoning processes are regarded as various series of rules, which are represented by a set of multiple vectors of T-invariant. Using these vectors, the algorithm for detecting what kind of errors exists in the reasoning processes is derived, where the structural errors of the rule set considered here are the series of rules which should not exist on the rule set, and are indispensable to be searched, detected and corrected.

A Multi-layered Planning Architecture for Soccer Agent

Based on manual simulation experiments, we propose a type of agent as a MultiLayerd Planning(MLP) Architecture to identify the game situation and create action policy. As a team, we will arrange the various type of agent according to game strategy, and realize semi-cooperative Multiagent model with minimum amount of communication.

A step towards a new test for learnability of machine learning

Based on the PAC (probably approximately correct) learning, a new test for learnability is proposed from the viewpoint of rate distortion theory. The criterion depends on the potential property of concept classes, which shows the relationship between sample complexity and accuracy.<<ETX>>