In real life, data are not always generated under stationary environments. However, traditional learning systems have normally assumed that the property of data streams is stationary over time, and this sometimes leads to the degradation in the system performance when there are some hidden contexts changes (e.g. changes in class boundaries and temporal trends in time series). Such context changes are called concept drifts, and various methods to handle concept drifts have been developed in machine learning and data mining fields. However, most of them are aiming for building classifier models. Considering that class boundaries have changed over time under non-stationary environments, extracted features should also be adapted to concept drifts autonomously. In this paper, we propose an extension of incremental linear discriminant analysis (ILDA) as an online feature extraction method under non-stationary environments. The extended ILDA has the following two functions: concept-drift detection and knowledge transfer. The recognition performance of the extended ILDA is evaluated for three benchmark data sets. Experimental results demonstrate that the recognition performance in the extended ILDA is greatly improved by introducing the knowledge transfer after the concept-drift detection.
This paper is concerned with construction of non-parametric PWA(Piecewise Affine) models using ℓ1 optimization. Though its effectiveness was demonstrated in the literature, the number of modes in the models tends to increase over ten thousands. Hence it is difficult to apply the existing control methods. This paper proposes a method to reduce the modes of this model by defining the significance of data based on ℓ1 optimization. Also by this method the size of data to represent this model becomes small. The performance of the method is shown through an application to the non-parametoric PWA model of motor.
This paper proposes human movement trajectory (HMT) extraction system and a state of people estimation system by thermopile array sensors. In our systems, sensors are attached at the ceiling and acquire thermal distribution, which are two-dimensional temperatures. The system distinguishes humans, object and others by fuzzy inference based on human characteristics, such as body temperature and movement. Each human is classied by the connected-component labeling. In the HMT extraction system, it extracts HMT as label centroids trajectory. In the state of people estimation system, it distinguishes adjoining people based on shape of human distribution in label image and estimates the number of human as the number of labels. In the HMT extraction experiment, we employed an adult and he performed 15 movements. As the results, the system successfully extracted HMTs with 78[%] accuracy and positional error was 21.5[cm]. In the state estimation experiment, we employed 4 adults and they performed 4 movements. As the results, the system successfully estimated the number of humans with 52[%] accuracy.
For the design of spacecraft transfer orbits in the Earth-Moon system, we need to model the dynamics in the context of the Planar Elliptic Restricted Three-Body Problem (PER3BP) since it has a non-negligible eccentricity. In this paper, we investigate invariant manifolds of the PER3BP by analyzing Lagrangian Coherent Structures (LCS). In particular, we show that the transfer orbits from the exterior realm to Moon can be developed by employing the geometric characteristics of the LCS obtained by long time numerical integrations.
In this paper, we propose a novel learning method which can estimate self-location of a robot and concepts of location simultaneously. A robot performs a probabilistic self-localization from sensor data. We integrate ambiguous speech recognition results with the model for self-localization on Bayesian approach. Experimental results show that a robot can obtain words for several locations and make use of them in self-localization task. In addition, we evaluate the performance of lexical acquisition task about words for places and show its effectiveness.
It is general to evaluate the strength of a structure by the strain and the stress obtained by using the finite element method. The strong shape can be derived for an overload when these evaluations are used. However, it is not an enough evaluation if it thinks from the viewpoint used in a mid/long term. In this paper, we propose the damage estimation algorithm which estimates the damage of the structure by using the structural analysis and the linear cumulative damage law. Finally, this proposal method is developed to the optimization problem, and we construct the optimization method of practical use which can design strong shapes corresponding to cumulative fatigue while estimating damage. The effectiveness of this proposed method is shown by applying to the shape optimization problem of the plunger tip in the die-cast process.
In this paper, we discuss the implementation of a control system using reference governor for the control of constraint systems. This control scheme uses the maximal output admissible set to determine the reference input r[k] at the time k. When the number of facets of the maximal output admissible set is very large, it might happen that the computing time to determine r[k] exceeds the sampling period. To reduce the computing time, we propose an efficient scheme to determine r[k] using the information of the facet Fik−1 which is used to determine r[k−1] and the set of facets which are adjacent to Fik−1 . Set of adjacent facets for every facet of the maximal output admissible set are computed in advance. Moreover, the effectiveness of proposing method is verified by simulation and experiment for a cart system with pendulum.