Embedding new observations via sparse-coding for non-linear manifold learning

Non-linear dimensionality reduction techniques are affected by two critical aspects: (i) the design of the adjacency graphs, and (ii) the embedding of new test data—the out-of-sample problem. For the first aspect, the proposed solutions, in general, were heuristically driven. For the second aspect, the difficulty resides in finding an accurate mapping that transfers unseen data samples into an existing manifold. Past works addressing these two aspects were heavily parametric in the sense that the optimal performance is only achieved for a suitable parameter choice that should be known in advance.     

Design and validation of a motion-tracking system for ROM measurements in home rehabilitation

This work deals with the design, implementation and experimental validation of a home rehabilitation system capable of bringing a real-time 3D reconstruction of human posture to the therapist’s desk. The system has been validated against two commercial systems, capable of providing joint angle measurements and commonly used in rehabilitation gyms, which are used profitably to assist clinics in limb motor dysfunction treatment. Early experimental results promise to provide measurements accurate enough to complement the subjective therapist’s perception with objective observations, leading to quantitative evaluation.     

Stock trend prediction based on a new status box method and AdaBoost probabilistic support vector machine

Stock trend prediction is regarded as one of the most challenging tasks of financial time series prediction. Conventional statistical modeling techniques are not adequate for stock trend forecasting because of the non-stationarity and non-linearity of the stock market. With this regard, many machine learning approaches are used to improve the prediction results. These approaches mainly focus on two aspects: regression problem of the stock price and prediction problem of the turning points of stock price. In this paper, we concentrate on the evaluation of the current trend of stock price and the prediction of the change orientation of the stock price in future. Then, a new approach named status box method is proposed. Different from the prediction issue of the turning points, the status box method packages some stock points into three categories of boxes which indicate different stock status. And then, some machine learning techniques are used to classify these boxes so as to measure whether the states of each box coincides with the stock price trend and forecast the stock price trend based on the states of the box. These results would support us to make buying or selling strategies. Comparing with the turning points prediction that only considered the features of one day, each status box contains a certain amount of points which represent the stock price trend in a certain period of time. So, the status box reflects more information of stock market. To solve the classification problem of the status box, a special features construction approach is presented. Moreover, a new ensemble method integrated with the AdaBoost algorithm, probabilistic support vector machine (PSVM), and genetic algorithm (GA) is constructed to perform the status boxes classification. To verify the applicability and superiority of the proposed methods, 20 shares chosen from Shenzhen Stock Exchange (SZSE) and 16 shares from National Association of Securities Dealers Automated Quotations (NASDAQ) are applied to perform stock trend prediction. The results show that the status box method not only have the better classification accuracy but also effectively solve the unbalance problem of the stock turning points classification. In addition, the new ensemble classifier achieves preferable profitability in simulation of stock investment and remarkably improves the classification performance compared with the approach that only uses the PSVM or back-propagation artificial neural network (BPN).     

Deep Scalogram Representations for Acoustic Scene Classification

Spectrogram representations of acoustic scenes have achieved competitive performance for acoustic scene classification. Yet, the spectrogram alone does not take into account a substantial amount of time-frequency information. In this study, we present an approach for exploring the benefits of deep scalogram representations, extracted in segments from an audio stream. The approach presented firstly transforms the segmented acoustic scenes into bump and morse scalograms, as well as spectrograms; secondly, the spectrograms or scalograms are sent into pre-trained convolutional neural networks; thirdly, the features extracted from a subsequent fully connected layer are fed into (bidirectional) gated recurrent neural networks, which are followed by a single highway layer and a softmax layer; finally, predictions from these three systems are fused by a margin sampling value strategy. We then evaluate the proposed approach using the acoustic scene classification data set of 2017 IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE). On the evaluation set, an accuracy of 64.0% from bidirectional gated recurrent neural networks is obtained when fusing the spectrogram and the bump scalogram, which is an improvement on the 61.0% baseline result provided by the DCASE 2017 organisers. This result shows that extracted bump scalograms are capable of improving the classification accuracy, when fusing with a spectrogram-based system.      

Extension of some MAP results to transient MAPs and Markovian binary trees

In this work we extend previous results on moment-based characterization and minimal representation of stationary Markovian arrival processes (MAPs) and rational arrival processes (RAPs) to transient Markovian arrival processes (TMAPs) and Markovian binary trees (MBTs).     

New Uni-Resonance Approximation Technique for Reaction Rate of Higher Resonances in Activation Detectors

As a high efficiency solid sample analysis technique based on resonance ionization spectroscopy (RIS). Resonant Laser Ablation (RLA) is available. RLA combines Laser Ablation (LA) and RIS simultaneously with a single laser device. We have proposed the new dosimetry technique for a solid sample based on RLA.

We have attempted the detection of ²⁶Al (T _1/2: 7.2×10⁵yr) produced in the high purity Al irradiated by 14MeV neutrons at Fusion Neutronics Source (FNS). Japan Atomic Energy Research Institute. Some interference effects are observed in the trace element detection experiments. These interference effects interrupt the detection of ²⁶Al. In conclusion. it is difficult that ²⁶Al is recognized clearly. We have mainly discussed the origins and elimination methods of these interference effects to detect the trace elements produced by nuclear reactions.     

The method of separation for evolutionary spectral density estimation of multi-variate and multi-dimensional non-stationary stochastic processes

The method of separation can be used as a non-parametric estimation technique, especially suitable for evolutionary spectral density functions of uniformly modulated and strongly narrow-band stochastic processes. The paper at hand provides a consistent derivation of method of separation based spectrum estimation for the general multi-variate and multi-dimensional case. The validity of the method is demonstrated by benchmark tests with uniformly modulated spectra, for which convergence to the analytical solution is demonstrated. The key advantage of the method of separation is the minimization of spectral dispersion due to optimum time- or space–frequency localization. This is illustrated by the calibration of multi-dimensional and multi-variate geometric imperfection models from strongly narrow-band measurements in I-beams and cylindrical shells. Finally, the application of the method of separation based estimates for the stochastic buckling analysis of the example structures is briefly discussed.