A modified conjugate gradient-based Elman neural network

Elman recurrent network is a representative model with feedback mechanism. Although gradient descent method has been widely used to train Elman network, it frequently leads to slow convergence. According to optimization theory, conjugate gradient method is an alternative strategy in searching the descent direction during training. In this paper, an efficient conjugate gradient method has been presented to reach the optimal solution in two ways: (1) constructing a more effective conjugate coefficient, (2) determining adaptive learning rates in terms of the generalized Armijo search method. Experiments show that the performance of the new algorithm is superior to traditional algorithms, such as gradient descent method and conjugate gradient method. In particular, the new algorithm has better performance than the evolutionary algorithm. Finally, we prove the weak and strong convergence of the presented algorithm, i.e., the gradient norm of the error function with respect to the weight vectors converges to zero and the weight sequence approaches a fixed optimal point.     

Evaluating pedestrian interaction preferences with a game theoretic autonomous vehicle in virtual reality

Localisation and navigation of autonomous vehicles (AVs) in static environments are now solved problems, but how to control their interactions with other road users in mixed traffic environments, especially with pedestrians, remains an open question. Recent work has begun to apply game theory to model and control AV-pedestrian interactions as they compete for space on the road whilst trying to avoid collisions. But this game theory model has been developed only in unrealistic lab environments. To improve their realism, this study empirically examines pedestrian behaviour during road crossing in the presence of approaching autonomous vehicles in more realistic virtual reality (VR) environments. The autonomous vehicles are controlled using game theory, and this study seeks to find the best parameters for these controls to produce comfortable interactions for the pedestrians. In a first experiment, participants’ trajectories reveal a more cautious crossing behaviour in VR than in previous laboratory experiments. In two further experiments, a gradient descent approach is used to investigate participants’ preference for AV driving style. The results show that the majority of participants were not expecting the AV to stop in some scenarios, and there was no change in their crossing behaviour in two environments and with different car models suggestive of car and last-mile style vehicles. These results provide some initial estimates for game theoretic parameters needed by future AVs in their pedestrian interactions and more generally show how such parameters can be inferred from virtual reality experiments.     

Exploratory study of the effect of binaural beat stimulation on the EEG activity pattern in resting state using artificial neural networks

Anxiety disorders afflict almost 7.3 percent of the world’s population. One in 14 people will experience anxiety disorder at the given year. When associated with mood disorders, anxiety can also trigger or increase other diseases’ symptoms and effects, like depression and suicidal behavior. Binaural beats are a low-frequency type of acoustic stimulation perceived when the individual is subjected to two slightly different wave frequencies, from 200 to 900 Hz. Binaural beats can contribute to anxiety reduction and modification of other psychological conditions and states, modifying cognitive processes and mood states. In this work, we applied a 5 Hz binaural beat to 6 different subjects, to detect a relevant change in their brainwaves before and after the stimuli. We applied 20 min stimuli in 10 separated sessions. We assessed the differences using a Multi-Layer Perceptron classifier in comparison with non-parametric tests and Low-Resolution Brain Electromagnetic Tomography (eLORETA). eLORETA showed remarkable changes in High Alpha. Both eLORETA and MLP approaches revealed outstanding modifications in high Beta. MLP evinced significant changes in Theta brainwaves. Our study evidenced high Alpha modulation at the limbic lobe, implicating in a possible reduction of sympathetic system activation in the studied sample. Our main results on eLORETA suggest a strong increase in the current distribution, mostly in Alpha 2, at the Anterior Cingulate, which is related to the monitoring of mistakes regarding social conduct, recognition and expression of emotions. We also found that MLPs are able of evincing the main differences with high separability in Delta and Theta.     

Comparing eye-tracking metrics of mental workload caused by NDRTs in semi-autonomous driving

The objective of this study was to verify the effectiveness of eye-tacking metrics in indicating driver’s mental workload in semi-autonomous driving when the driver is engaged in different non-driving related tasks (NDRTs). A driving simulator was developed for three scenarios (high-, medium-, and low-mental workload presented by SAE (Society of Automotive Engineers) Levels 0, 1, and 2) and three uni-modality secondary tasks. Thirty-six individuals participated in the driving simulation experiment. NASA-TLX (Task Load Index), secondary task performance, and eye-tracking metrics were used as indicators of mental workload. The subjective rating using the NASA-TLX showed a main effect of autonomous level on mental workload in both visual and auditory tasks. Correlation-matrix calculation and principal-component extraction indicated that pupil diameter change, number of saccades, saccade duration, fixation duration, and 3D gaze entropy were effective indicators of a driver’s mental workload in the visual and auditory multi-tasking situations of semi-autonomous driving. The accuracy of predicting the mental-workload level using the K-Nearest Neighbor (KNN) classifier was 88.9% with bootstrapped data. These results can be used to develop an adaptive multi-modal interface that issues efficient and safe takeover requests.     

Derivative free gradient projection algorithms for rotation

A simple modification substantially simplifies the use of the gradient projection (GP) rotation algorithms of Jennrich (2001, 2002). These algorithms require subroutines to compute the value and gradient of any specific rotation criterion of interest. The gradient can be difficult to derive and program. It is shown that using numerical gradients gives almost precisely the same results as using exact gradients. The resulting algorithm is very easy to use because the only problem specific code required is that needed to define the rotation criterion. The computing time is increased when using numerical gradients, but it is still very modest for most purposes. While used extensively elsewhere, numerical derivatives seem to be underutilized in statistics.     

Fixation identification in centroid versus start-point modes using eye-tracking data

Fixation-identification algorithms, needed for analyses of eye movements, may typically be separated into three categories, viz. (i) velocity-based algorithms, (ii) area-based algorithms, and (iii) dispersion-based algorithms. Dispersion-based algorithms are commonly used but this application introduces some difficulties, one being optimization. Basically, there are two modes to reach this goal of optimization, viz., the start-point mode and the centroid mode. The aim of the present study was to compare and evaluate these two dispersion-based algorithms. Manual inspections were made of 1,400 fixations in each mode. Odds ratios showed that by using the centroid mode for fixation detection, a valid fixation is 2.86 times more likely to be identified than by using the start-point mode. Moreover, the algorithm based on centroid mode dispersion showed a good interpretation speed, accuracy, robustness, and ease of implementation, as well as adequate parameter settings.     

How experimental algorithmics can benefit from Mayo’s extensions to Neyman–Pearson theory of testing

Although theoretical results for several algorithms in many application domains were presented during the last decades, not all algorithms can be analyzed fully theoretically. Experimentation is necessary. The analysis of algorithms should follow the same principles and standards of other empirical sciences. This article focuses on stochastic search algorithms, such as evolutionary algorithms or particle swarm optimization. Stochastic search algorithms tackle hard real-world optimization problems, e.g., problems from chemical engineering, airfoil optimization, or bio-informatics, where classical methods from mathematical optimization fail. Nowadays statistical tools that are able to cope with problems like small sample sizes, non-normal distributions, noisy results, etc. are developed for the analysis of algorithms. Although there are adequate tools to discuss the statistical significance of experimental data, statistical significance is not scientifically meaningful per se. It is necessary to bridge the gap between the statistical significance of an experimental result and its scientific meaning. We will propose some ideas on how to accomplish this task based on Mayo’s learning model (NPT*).     

An optimization approach for mapping and measuring the divergence and correspondence between paths

Many domains of empirical research produce or analyze spatial paths as a measure of behavior. Previously, approaches for measuring the similarity or deviation between two paths have either required timing information or have used ad hoc or manual coding schemes. In this paper, we describe an optimization approach for robustly measuring the area-based deviation between two paths we call ALCAMP (Algorithm for finding the Least-Cost Areal Mapping between Paths). ALCAMP measures the deviation between two paths and produces a mapping between corresponding points on the two paths. The method is robust to a number of aspects in real path data, such as crossovers, self-intersections, differences in path segmentation, and partial or incomplete paths. Unlike similar algorithms that produce distance metrics between trajectories (i.e., paths that include timing information), this algorithm uses only the order of observed path segments to determine the mapping. We describe the algorithm and show its results on a number of sample problems and data sets, and demonstrate its effectiveness for assessing human memory for paths. We also describe available software code written in the R statistical computing language that implements the algorithm to enable data analysis.     

Learning Continuous Probability Distributions with Symmetric Diffusion Networks

In this article we present symmetric diffusion networks, a family of networks that instantiate the principles of continuous, stochastic, adaptive and interactive propagation of information. Using methods of Markovion diffusion theory, we formalize the activation dynamics of these networks and then show that they can be trained to reproduce entire multivariate probability distributions on their outputs using the contrastive Hebbion learning rule (CHL). We show that CHL performs gradient descent on an error function that captures differences between desired and obtained continuous multivariate probability distributions. This allows the learning algorithm to go beyond expected values of output units and to approximate complete probability distributions on continuous multivariate activation spaces. We argue that learning continuous distributions is an important task underlying a variety of real-life situations that were beyond the scope of previous connectionist networks. Deterministic networks, like back propagation, cannot learn this task because they are limited to learning average values of independent output units. Previous stochastic connectionist networks could learn probability distributions but they were limited to discrete variables. Simulations show that symmetric diffusion networks can be trained with the CHL rule to approximate discrete and continuous probability distributions of various types.     

相生相成：拥有生命意义与亲社会行为相互关系的纵向研究

本研究旨在探讨拥有生命意义与亲社会行为的相互关系。研究1对我国10省15所高校2375名学生的拥有生命意义、亲社会行为进行横断面调查。研究2对四川某高校878名学生的拥有生命意义和亲社会行为进行间隔三个月和六个月的三次追踪测量。结果发现：大学生拥有生命意义与亲社会行为具有相互促进的关系,拥有生命意义能稳定正向预测亲社会行为,亲社会行为也能够在一定程度上促进拥有生命意义,但这种促进作用不够稳定;拥有生命意义对利他的、情绪的、匿名的、依从的、紧急的亲社会行为有显著正向影响;而只有利他的亲社会行为能显著预测拥有生命意义。     

An improved stochastic EM algorithm for large-scale full-information item factor analysis

In this paper, we explore the use of the stochastic EM algorithm (Celeux & Diebolt (1985) Computational Statistics Quarterly, 2, 73) for large-scale full-information item factor analysis. Innovations have been made on its implementation, including an adaptive-rejection-based Gibbs sampler for the stochastic E step, a proximal gradient descent algorithm for the optimization in the M step, and diagnostic procedures for determining the burn-in size and the stopping of the algorithm. These developments are based on the theoretical results of Nielsen (2000, Bernoulli, 6, 457), as well as advanced sampling and optimization techniques. The proposed algorithm is computationally efficient and virtually tuning-free, making it scalable to large-scale data with many latent traits (e.g. more than five latent traits) and easy to use for practitioners. Standard errors of parameter estimation are also obtained based on the missing-information identity (Louis, 1982, Journal of the Royal Statistical Society, Series B, 44, 226). The performance of the algorithm is evaluated through simulation studies and an application to the analysis of the IPIP-NEO personality inventory. Extensions of the proposed algorithm to other latent variable models are discussed.     

Driving manoeuvre during lane maintenance in older adults: Associations with neuropsychological scores

Older drivers experience difficulties in lane maintenance under challenging driving sections due to age-related cognitive declines, yet there is little comprehensive evidence on associations between cognitive functions and the lane maintenance in this population. In this study, fifty older drivers completed an on-road driving assessment and a battery of standard neuropsychological tests. Mean Lane Position (MLP), Standard Deviation of Lane Position (SDLP) and manoeuvre time calculated from precise vehicle movement trajectories were used as the lane maintenance parameters. The GNSS tracking vehicle movement presents comprehensive and reliable vehicle position data, which is more sensitive for detecting subtle variations of lane maintenance in older drivers. Statistical analysis results showed that lower visual attention (selective and divided attention) was associated with higher MLP and SDLP; MLP was also correlated to spatial abilities, executive function, and motor speed; manoeuvre time was negatively correlated with drivers’ risk-taking personality (all p < .01). Selective attention was found to be the best predictor of MLP in lane maintenance. A combined eight variables from three neuropsychological tests, UFOV 2 and 3, BD and BJLO, D-KEFS TMT 1, 2, 3, and 4, correctly classified 80.4% of participants with good versus low-performing lane maintenance.     

Riemannian Newton and trust-region algorithms for analytic rotation in exploratory factor analysis

In exploratory factor analysis, latent factors and factor loadings are seldom interpretable until analytic rotation is performed. Typically, the rotation problem is solved by numerically searching for an element in the manifold of orthogonal or oblique rotation matrices such that the rotated factor loadings minimize a pre-specified complexity function. The widely used gradient projection (GP) algorithm, although simple to program and able to deal with both orthogonal and oblique rotation, is found to suffer from slow convergence when the number of manifest variables and/or the number of latent factors is large. The present work examines the effectiveness of two Riemannian second-order algorithms, which respectively generalize the well-established truncated Newton and trust-region strategies for unconstrained optimization in Euclidean spaces, in solving the rotation problem. When approaching a local minimum, the second-order algorithms usually converge superlinearly or even quadratically, better than first-order algorithms that only converge linearly. It is further observed in Monte Carlo studies that, compared to the GP algorithm, the Riemannian truncated Newton and trust-region algorithms require not only much fewer iterations but also much less processing time to meet the same convergence criterion, especially in the case of oblique rotation.     

The logic of social exchange: has natural selection shaped how humans reason? Studies with the Wason selection task

In order to successfully engage in social exchange--cooperation between two or more individuals for mutual benefit--humans must be able to solve a number of complex computational problems, and do so with special efficiency. Following Marr (1982), Cosmides (1985) and Cosmides and Tooby (1989) used evolutionary principles to develop a computational theory of these adaptive problems. Specific hypotheses concerning the structure of the algorithms that govern how humans reason about social exchange were derived from this computational theory. This article presents a series of experiments designed to test these hypotheses, using the Wason selection task, a test of logical reasoning. Part I reports experiments testing social exchange theory against the availability theories of reasoning; Part II reports experiments testing it against Cheng and Holyoak's (1985) permission schema theory. The experimental design included eight critical tests designed to choose between social exchange theory and these other two families of theories; the results of all eight tests support social exchange theory. The hypothesis that the human mind includes cognitive processes specialized for reasoning about social exchange predicts the content effects found in these experiments, and parsimoniously explains those that have already been reported in the literature. The implications of this line of research for a modular view of human reasoning are discussed, as well as the utility of evolutionary biology in the development of computational theories.     

COMPARISON OF STAFF TRAINING STRATEGIES TO PROMOTE GENERALIZED TEACHING SKILLS

Two studies compared the effectiveness of different strategies for promoting generalization of staff skills in teaching self-care routines to clients with developmental disabilities. In Study 1, 9 direct-care staff members of group homes were trained sequentially through four conditions: (a) the provision of written instructions, (b) performance-based training using a single client program exemplar and simulated clients (single case training), (c) performance-based training using actual developmentally delayed clients as trainees (common stimuli training), and (d) performance-based training using multiple client program exemplars with simulated clients (general case training). The results indicated that staff members did not reach all generalization criteria until general case training was provided. Because staff members had been trained sequentially through several conditions in Study 1, a second study controlled for potential sequence effects. In Study 2, 7 staff members were trained using only the general case strategy after baseline. All staff members reached generalization criteria with only general case training, replicating the findings of Study 1. Together, the two studies demonstrated that the general case training strategy was more effective at promoting generalized training effects across clients, settings, and client programs than other commonly used staff training approaches.     

Effects of variable-interval value and amount of training on stimulus generalization.

In Experiment 1 pigeons pecked a key that was illuminated with a 501-nm light and obtained food by doing so according to a variable-interval (VI) schedule of reinforcement, the mean value of which differed across groups: either 30 s, 120 s, or 240 s. The pigeons in all three groups were trained for 10 50-min sessions. Generalization testing was conducted in extinction with different wavelengths of light. Absolute and relative generalization gradients were similar in shape for the three groups. Experiment 2 was a systematic replication of Experiment 1 using line orientation as the stimulus dimension and a mean VI value of either 30 s or 240 s. Again, gradients of generalization were similar for the two groups. In Experiment 3 pigeons pecked a key that was illuminated with a 501-nm light and obtained food reinforcers according to either a VI 30-s or a 240-s schedule. Training continued until response rates stabilized (> 30 sessions). For subjects trained with the 30-s schedule, generalization gradients were virtually identical regardless of whether training was for 10 sessions (Experiment 1) or until response rates stabilized. For subjects trained with the VI 240-s schedule, absolute generalization gradients for subjects trained to stability were displaced upward relative to gradients for subjects trained for only 10 sessions (Experiment 1), and relative generalization gradients were slightly flatter. These results indicate that the shape of a generalization gradient does not necessarily depend on the rate of reinforcement during 10-session single-stimulus training but that the effects of prolonged training on stimulus generalization may be schedule dependent.     

Generalization of free-operant avoidance behavior in pigeons

Three groups of four pigeons, trained to press a treadle on a free-operant avoidance schedule, were given auditory discrimination training. Alternating 2-min components of avoidance and no shock were paired with either a tone or white noise. The pigeons were subsequently given two types of generalization tests, with and without avoidable shocks scheduled. Two of the groups, trained interdimensionally, produced excitatory and inhibitory generalization gradients along the tone frequency dimension. A predicted post-discrimination gradient was computed from the algebraic summation of these gradients of excitation and inhibition. The predicted gradient was compared with the actual post-discrimination gradient obtained from the third group of pigeons that had been given intradimensional discrimination training on the tone frequency dimension. The predicted postdiscrimination gradient agreed in shape with the empirical postdiscrimination gradient. The results in general support Spence's (1937) gradient interaction theory.