Adapting to various types of concept drift is important for dealing with real-world online learning problems. To achieve this, we previously reported an system that uses ensemble classifiers, the adaptive classifiers-ensemble (ACE) system. ACE consists one classifier, many batch and a detection mechanism. improve performance ACE, have improved weighting method, which combines outputs added new classifier pruning method. Experimental results showed enhanced performed well synthetic dataset...

A calibration method for a structured light system by observing planar object from unknown viewpoints is proposed. captures 3D shape camera that observes stripe on an illuminated projector. The shape, obtained the defined pinhole model projection of stripe, solved using equation plane coefficients each stripe¿s are estimated 4×3 image-to-camera transformation matrix expressed parameters. Experimental results demonstrate high degree accuracy when following proposed approach.

It has been challenging to recognize walking humans at arbitrary poses from a single or small number of video cameras. Attempts have made mostly using 2D image/silhouette-based representation and limited use 3D kinematic model-based approaches. In this paper, the problem recognizing is addressed. Unlike all previous work in computer vision pattern recognition models are built sensed range data selected without any markers. An instance individual different pose recognized that pose. Both...

Concept drift, the change over time of statistical properties target variable, is a serious problem for online learning systems. To overcome this problem, we propose method inspired by human behavior detecting sudden concept drift. We first conducted experiment to investigate our working hypothesis that humans can detect changes quickly when their confident classifications are rejected despite fact recent classification accuracy high. The experiments supported hypothesis. then have proposed...

The demand for learning machines that can adapt to concept change, the change over time of statistical properties a target variable, has become more urgent. We, therefore, propose system in which multiple online and offline classifiers are used changing concepts. Our is able to: respond both sudden gradual changes, handle recurring concepts, detect occurrence understand hidden contexts past predict next concept. We evaluate effectiveness our system's elements demonstrate performed well with...

Recent improvements in embedded systems has enabled learning algorithms to provide realistic solutions for system identification problems. Existing algorithms, however, continue have limitations on systems, where physical memory space is constrained. To overcome this problem, we propose a Limited General Regression Neural Network (LGRNN), which variation of general regression neural network proposed by Specht or simplified fuzzy inference systems. The LGRNN continues incremental even if the...

Abstract The effect of applied strain on martensitic transformation in a superelastic Ti 46.4 Ni 47.6 Nb 6.0 alloy at room temperature was investigated by tensile tests, differential scanning calorimetry measurements, and X‐ray diffraction. Reverse starting ( A s ) finishing f temperatures increased with the application tensile‐strain over 13%, undeformed specimen showing = −29.2°C 17.9°C, while 13% predeformed exhibited 37.1°C 40.2°C. Furthermore, values for almost recovered to those when...

Driver drowsiness is a widely recognized cause of motor vehicle accidents. Therefore, reduction in drowsy driving crashes required. Many studies evaluating the crash risk and developing detection systems, have used observer rating (ORD) as reference standard (i.e. ground truth) drowsiness. ORD method human raters levels driver drowsiness, by visually observing driver. Despite widespread use ORD, concerns remain regarding its convergent validity, which supported relationship between other...

In this paper, we propose a limited general regression neural network (LGRNN) for embedded systems. The LGRNN is an improved version of that continues incremental learning under fixed number hidden units. Initially, the learns new samples incrementally by allocating If units reaches upper bound, has to remove one useless unit learn sample. However, there are cases in which adverse effects removing greater than positive case, should refrain from To achieve this, predicts several options...

Photo Voltaic (PV) devices have a Maximum Power Point (MPP) at which they generate maximum power. Because the MPP depends on solar radiation and PV panel temperature, it is not constant over time. A Tracker (MPPT) widely used to continuously obtain power, but if changes rapidly, efficiency of most classic MPPT (e.g., Perturbation Observation (P&O) method) reduces. controllers using neural network respond quickly rapidly changing must usually undergo prelearning PV-specific data, so we...

A compact and high-speed 3D human body measurement system is proposed. Whole data which we can take from systems that have been developed with a few viewpoints not successfully acquired due to occlusion. It proposed method obtain successful by allocating multiple rangefinders correctly. Four are installed in pole. Those four pole units 16 assigned around human. Multiple viewpoint range images allow the shape reconstruction for body. The time 2 seconds average error found be 1.88 mm. In this...

The model selection for neural networks is an essential procedure to get not only high levels of generalization but also a compact data model. Especially in terms getting the model, usually outperform other kinds machine learning methods. Generally, models are selected by trial and error testing using whole samples given advance. In many cases, however, it difficult time consuming prepare To overcome these inconveniences, we propose hybrid on-line system radial basis function (RBF) network...

Nitinol shape memory alloys (SMAs) are attracting considerable attention as core materials for medical guidewires because of their excellent flexibility and retention. However, since possess low rigidity, the pushability torquability insufficient. On other hand, although made stainless steel have high pushability, plastic deformation occurs easily. We developed a new class superelastic with functionally graded properties from tip to end by using SMA such Cu-Al-Mn-based or Ni-free Ti-Mo-Sn...

This paper presents an incremental learning and model selection method under the virtual concept drifting environments, where their prior distribution of inputs is changing over time. In previous work, a statistical drift was constructed, model-selection criterion for radial basis function neural networks (RBFNNs) such environments built with environmental (Yamauchi 2009). However, in model, no consideration given to reducing computational complexity storage space storing learned samples...