In this paper, information incompleteness in tabular data and rule generation in tabular data are focused on, and the previously proposed important researches are surveyed from the viewpoint of rough sets and data mining. Then, “Rough sets Non-deterministic Information Analysis (RNIA)” and “a NIS-Apriori system” proposed by the authors are described. The framework of rule generation from tables with information incompleteness is also outlined. RNIA offers a new framework of rough sets based on possible world semantics, and it gives one solution algorithm for the rule generation problem considering information incompleteness. This solution algorithm is termed “NIS-Apriori” and implemented as a NIS-Apriori system in SQL language. We also mention the new possibility of rough sets non-deterministic information analysis that employs the NIS-Apriori algorithm as the core algorithm.
In this paper, we improve Takahashi et al.’s method for extracting pseudo-generalized dynamic reducts (pGDRs) from a decision table with numerous objects and attributes. Takahashi et al.’s method consists of pGDR candidates extraction phase and pGDR confirmation phase using training datasets. However, a parameter ε used in the confirmation phase is required to set appropriately before starting the confirmation phase. Moreover, it is difficult to interrupt the confirmation processes for a pGDR candidate G even though it is expected that G does not satisfy the condition of pGDR. To solve these two issues, a dynamic update method of the parameter ε and an interruption method of the confirmation processes based on binomial test are introduced to the confirmation phase. Moreover, robustness of the extracted pGDRs to test datasets is examined.
In Epistemic Situation Calculus (ES) proposed by Lakemeyer and Levesque, by assuming a situation as a possible world, it is possible to interpret an action as a kind of modality. Moreover, since the state of knowledge of the agent is interpreted by the equivalence relation to the world after the action, knowledge representation based on granulation is possible. In this paper, we apply granular reasoning to the epistemic situation calculus by interpreting actions as modalities and granules of possible worlds as states. The zoom reasoning proposed by Murai et al. is regarded as a cognitive action and is incorporated into the ES as an abstraction and refinement action by the granularity of the situation. The relationship between rough sets and semantic interpretation based on Belnap’s four-valued logic is given as a model of ES, and a model of ES with possible worlds and four-valued logic is presented.
The document classification task is a well-known task for natural language processing. In this paper, I propose a Rough Set Theory based document classification system. First, the proposed system makes a decision table by combining the label of the document and terms extracted by the document frequency and reduction. Next, it extracts decision rules from upper approximation and lower approximation, respectively. Then it matches an unlabeled document to both decision rules and extracts a label which has the maximum value of the sum of rules’ weight. I use SI (Satisfaction Index), CI (Coverage Index) and Lift as the rules’ weight. In order to evaluate this approach, I implemented a prototype system and tried to classify labeled patent publications in Japanese with experts. This system could extract some rules evaluated as useful by an expert and shows its accuracy rate is higher than by selecting the modal label. However, the rate of the useful rules is only 25% and the accuracy rate and the Kappa statistics are not enough to use. This result cannot also say this approach is better than Naive Bayes Classifier. In the next study, I improve this approach based on the analysis of this evaluation.
We consider several nonlinear integrals with respect to the monotone measures. We term the integrals defined by certain approximations via simple functions as decomposition type integrals. This category is classified based on the approximation directions (from above/below) and the disjointness of corresponding measurable set families (partition/covering). Furthermore, we add two classification points: finiteness of the summation (finite/countable) and sign of coefficients (non-negative/signed). This study aims to clarify the essential features of these integrals by considering several convergence theorems for each group of the decomposition type integrals.
This study aims to discuss and examine possibility of aggregation methods based on tools in cooperative game and evidence theory in a group decision making. In this paper, multiattribute data represented by evaluation vectors are extended toward a high dimensional space through a transformation introduced by Takahagi, a set function-representation of vectors. Some methods (methods for combination of evidence) in evidence theory are applied to aggregate evaluation data represented/extended as set functions. Some methods (solutions of game) in cooperative game theory are used as dimension reduction methods to represent set function form evaluation data as evaluation vectors.
In this article, a new approach for problem solving is proposed. The approach is named “Ka-Ki-Ku-Ke-Ko” loop. It consists of five stages that rhyme with “Ka,” “Ki,” “Ku,” “Ke,” and “Ko” in Japanese. The five stages are described separately. Some examples are shown to understand each stages and those whole process.
A method of a fusion of fuzzy inference and policy gradient reinforcement learning has been proposed that directly learns, as maximizes the expected value of the reward per episode, parameters in a policy function represented by fuzzy rules with weights. A study has applied this method to a task of speed control of an automobile and has obtained correct policies, some of which control speed of the automobile appropriately but many others generate inappropriate vibration of speed. In general, the policy is not desirable that causes sudden time change or vibration in the output value, and there would be many cases where the policy giving smooth time change in the output value is desirable. In this paper, we propose a fusion method using the objective function, that introduces defuzzification with the center of gravity model weighted stochastically and a constraint term for smoothness of time change, as an improvement measure in order to suppress sudden change of the output value of the fuzzy controller. Then we show the learning rule in the fusion, and also consider the effect by reward functions on the fluctuation of the output value. As experimental results of an application of our method on speed control of an automobile, it was confirmed that the proposed method has the effect of suppressing the undesirable fluctuation in time-series of the output value. Moreover, it was also showed that the difference between reward functions might adversely affect the results of learning.
Adversarial examples can be used to exploit vulnerabilities in neural networks and threaten their sensitive applications. Adversarial attacks are evolving daily, and are rapidly rendering defense methods that assume specific attacks obsolete. This paper proposes a new defense method that does not assume a specific adversarial attack, and shows that it can be used efficiently to protect a network from a variety of adversarial attacks. Adversarial perturbations are small values; consequently, an image quality recovery method is considered to be an effective way to remove adversarial perturbations because such a method often includes a smoothing effect. The proposed method, called the denoising-based perturbation removal network (DPRNet), aims to eliminate perturbations generated by an adversarial attack for image classification tasks. DPRNet is an encoder–decoder network that excludes adversarial images during training and can reconstruct a correct image from an adversarial image. To optimize DPRNet’s parameters for eliminating adversarial perturbations, we also propose a new perturbation removal loss (PRloss) metric, which consists of a reconstructed loss and a Kullback–Leibler divergence loss that expresses the class probability distribution difference between an original image and a reconstructed image. To remove adversarial perturbation, the proposed network is trained using various types of distorted images considering the proposed PRloss metric. Thus, DPRNet eliminates image perturbations, allowing the images to be classified easily. We evaluate the proposed method using the MNIST, CIFAR-10, SVHN, and Caltech 101 datasets and show that the proposed defense method invalidates 99.8%, 95.1%, 98.7%, and 96.0% of the adversarial images that are generated by several adversarial attacks in the MNIST, CIFAR-10, SVHN, and Caltech 101 datasets, respectively.
Studies on tracking athletes by image processing are being made for the purpose of utilizing data in sports using ICT. However, it is impossible to correctly obtain the position information of athletes in the types of games such as American Football in which occlusions occur frequently due to the athletes’ close formation on the field. Therefore, it is a problem that the moving track of an athletes is divided, which makes it difficult to make accurate tracking. In this research, we propose a tracking method for an athlete who is robust against occlusion by using the moving tracks obtained from respective video images shot from multiple points to make them complement each other. Using this, we aim at tracking in regular games to obtain information including that about the opponents that cannot be obtained from the device with GNSS sensor.