Generative and discriminative learning by CL-Net.

This correspondence presents a two-stage classification learning algorithm. The first stage approximates the class-conditional distribution of a discrete space using a separate mixture model, and the second stage investigates the class posterior probabilities by training a network. The first stage explores the generative information that is inherent in each class by using the Chow-Liu (CL) method, which approximates high-dimensional probability with a tree structure, namely, a dependence tree, whereas the second stage concentrates on discriminative learning to distinguish between classes. The resulting learning algorithm integrates the advantages of both generative learning and discriminative learning. Because it uses CL dependence-tree estimation, we call our algorithm CL-Net. Empirical tests indicate that the proposed learning algorithm makes significant improvements when compared with the related classifiers that are constructed by either generative learning or discriminative learning.     

Shadow-driven 4D haptic visualization

Just as we can work with two-dimensional floor plans to communicate 3D architectural design, we can exploit reduced-dimension shadows to manipulate the higher-dimensional objects generating the shadows. In particular, by taking advantage of physically reactive 3D shadow-space controllers, we can transform the task of interacting with 4D objects to a new level of physical reality. We begin with a teaching tool that uses 2D knot diagrams to manipulate the geometry of 3D mathematical knots via their projections; our unique 2D haptic interface allows the user to become familiar with sketching, editing, exploration, and manipulation of 3D knots rendered as projected imageson a 2D shadow space. By combining graphics and collision-sensing haptics, we can enhance the 2D shadow-driven editing protocol to successfully leverage 2D pen-and-paper or blackboard skills. Building on the reduced-dimension 2D editing tool for manipulating 3D shapes, we develop the natural analogy to produce a reduced-dimension 3D tool for manipulating 4D shapes. By physically modeling the correct properties of 4D surfaces, their bending forces, and their collisions in the 3D haptic controller interface, we can support full-featured physical exploration of 4D mathematical objects in a manner that is otherwise far beyond the experience accessible to human beings. As far as we are aware, this paper reports the first interactive system with force-feedback that provides "4D haptic visualization" permitting the user to model and interact with 4D cloth-like objects.     

Illuminance Flow Estimation by Regression

We investigate the estimation of illuminance flow using Histograms of Oriented Gradient features (HOGs). In a regression setting, we found for both ridge regression and support vector machines, that the optimal solution shows close resemblance to the gradient based structure tensor (also known as the second moment matrix).     

Frequency of the sit to stand task: An observational study of free-living adults

The sit to stand movement is a key determinant of functional independence. Knowledge of the frequency with which the sit to stand movement is performed throughout the day could inform workplace ergonomics, but has rarely been examined.Healthy adults (n = 140) were recruited from the general population. Free-living activity for each participant was reported using an activity monitor. On average, participants performed 60 (±22) sit to stand movements each day. Participants in indoor sedentary occupations performed significantly more sit to stand movements per day than participants in outdoor active occupations (66 vs. 54; n = 102; p = 0.003). Participants (n = 33) performed significantly more sit to stand movements on working days than on non-working days (65 vs. 55; p = 0.018).This analysis provides contemporary data for sit to stand frequency in a predominantly working population, and demonstrates that work and employment have a significant effect on that frequency.     

On-Line Shape recognition with incremental training using binary synaptic weights algorithm

Recognition of hand drawn shapes is beneficial in drawing packages and automated sketch entry in handheld computers. In this paper, we propose a new approach to on-line geometric shape recognition with incremental training function, which utilizes a heuristic function to reduce noise and a neural network for classification and on-line training. Instead of recognizing segments of a drawing and then performing syntactical analysis to match with a predefined shape, which is weak in terms of generalization and dealing with noise, we examine the shape as a whole. The main concept of the recognition method is derived from the fact that internal angles are very important in the perceived shape of outlines. Our application's aim is to recognize elliptic, rectangular, and triangular shapes in a way similar to human cognition of these shapes. Human beings recognize such basic shapes regardless of the variations in size, noise on the shape border, rotation and in the case of triangles, regardless of the type of the triangle. The key concept is that the neural network learns the relationships between the internal angles of a shape and its classification, therefore only a few training samples which represent the class of the shape are sufficient. Fast meremental training, which is performed on-line, is accomplished by the use of the Binary Synaptic Weights algorithm, a one pass, feedforward neural network training algorithm. Incremental training offers the advantage of adjusting the recognition capability of the system to the user's drawings. the results are very successful, such that the neural network correctly classified shapes that did not have any resemblance to the shapes in the initial training set.     

Linear Least-Squares Algorithms for Temporal Difference Learning

We introduce two new temporal difference (TD) algorithms based on the theory of linear least-squares function approximation. We define an algorithm we call Least-Squares TD (LS TD) for which we prove probability-one convergence when it is used with a function approximator linear in the adjustable parameters. We then define a recursive version of this algorithm, Recursive Least-Square TD (RLS TD). Although these new TD algorithms require more computation per time-step than do Suttons TD() algorithms, they are more efficient in a statistical sense because they extract more information from training experiences. We describe a simulation experiment showing the substantial improvement in learning rate achieved by RLS TD in an example Markov prediction problem. To quantify this improvement, we introduce the TD error variance of a Markov chain, TD, and experimentally conclude that the convergence rate of a TD algorithm depends linearly on TD. In addition to converging more rapidly, LS TD and RLS TD do not have control parameters, such as a learning rate parameter, thus eliminating the possibility of achieving poor performance by an unlucky choice of parameters.     

Numerical solution of a calculus of variations problem using the feedforward neural network architecture

It is demonstrated, through theory and numerical example, how it is possible to construct directly and noniteratively a feedforward neural network to solve a calculus of variations problem. The method, using the piecewise linear and cubic sigmoid transfer functions, is linear in storage and processing time. The L₂ norm of the network approximation error decreases quadratically with the piecewise linear transfer function and quartically with the piecewise cubic sigmoid as the number of hidden layer neurons increases. The construction requires imposing certain constraints on the values of the input, bias, and output weights, and the attribution of certain roles to each of these parameters.

All results presented used the piecewise linear and cubic sigmoid transfer functions. However, the noniterative approach should also be applicable to the use of hyperbolic tangents and radial basis functions.     

Graph structure in the Web

The study of the Web as a graph is not only fascinating in its own right, but also yields valuable insight into Web algorithms for crawling, searching and community discovery, and the sociological phenomena which characterize its evolution. We report on experiments on local and global properties of the Web graph using two AltaVista crawls each with over 200 million pages and 1.5 billion links. Our study indicates that the macroscopic structure of the Web is considerably more intricate than suggested by earlier experiments on a smaller scale.     

Development of a high‐resolution RGBW flexible display using a white organic light‐emitting diode with microcavity structure and that of a side‐roll touch panel

In this study, white organic electroluminescent devices with microcavity structures were developed. A flexible high‐resolution active‐matrix organic light‐emitting diode display with low power consumption using red, green, blue, and white sub‐pixels formed by a color‐filter method was fabricated. In addition, a side‐roll touch display was developed in combination with a capacitive flexible touch screen.     

Comparing techniques for authorship attribution of source code

Attributing authorship of documents with unknown creators has been studied extensively for natural language text such as essays and literature, but less so for non‐natural languages such as computer source code. Previous attempts at attributing authorship of source code can be categorised by two attributes: the software features used for the classification, either strings of n tokens/bytes (n‐grams) or software metrics; and the classification technique that exploits those features, either information retrieval ranking or machine learning. The results of existing studies, however, are not directly comparable as all use different test beds and evaluation methodologies, making it difficult to assess which approach is superior. This paper summarises all previous techniques to source code authorship attribution, implements feature sets that are motivated by the literature, and applies information retrieval ranking methods or machine classifiers for each approach. Importantly, all approaches are tested on identical collections from varying programming languages and author types. Our conclusions are as follows: (i) ranking and machine classifier approaches are around 90% and 85% accurate, respectively, for a one‐in‐10 classification problem; (ii) the byte‐level n‐gram approach is best used with different parameters to those previously published; (iii) neural networks and support vector machines were found to be the most accurate machine classifiers of the eight evaluated; (iv) use of n‐gram features in combination with machine classifiers shows promise, but there are scalability problems that still must be overcome; and (v) approaches based on information retrieval techniques are currently more accurate than approaches based on machine learning. Copyright © 2012 John Wiley & Sons, Ltd.