Normalized Prediction Error Research Articles

Tsunami Early Warning Using High-Frequency Ocean Radar System in the Kii Channel, Japan

Abstract The tsunami of the 2011 Tohoku earthquake was monitored by the high-frequency (HF) ocean radar system in the Kii Channel, Japan. In this study, we conducted a retrospective analysis of tsunami early warning based on HF radar observations. First, we synthesized the surface current velocity using observation data of two radar land stations: Minato and Saikazaki. After vector synthesis, the time series of current velocity showed high consistency with the results of forward simulation. Then, we used the time series starting from 08:00 (UTC), 11 March 2011, as the input for data assimilation, reconstructing the tsunami wavefield in the Kii Channel. Comparison with actual observations at Kobe showed that our approach can accurately forecast coastal tsunami waveforms at least 50 min prior to tsunami arrival. The normalized prediction error decreased over time, indicating a progressively improving forecast accuracy. This is the first study on tsunami data assimilation using HF radar for the early warning of an actual seismogenic tsunami. We also discussed the limitations of the current HF radar system in the Kii Channel and made suggestions regarding the deployment of future radar land stations.

Population pharmacokinetics of mycophenolate mofetil in pediatric patients early after liver transplantation.

Objective: To investigate the factors influencing the pharmacokinetics of mycophenolate mofetil (MMF) in pediatric patients after liver transplantation, and to establish a population pharmacokinetics model, which can provide a reference for clinical dosage adjustment. Methods: A prospective study in a single center was performed on pediatric patients who were administrated with mycophenolate mofetil dispersible tablets (MMFdt) for at least 4 days after liver transplantation continuously. Blood samples were collected in ethylene diamine tetraacetic acid anticoagulant tubes before dosing and 0.5, 1, 2, 4, 8, and 12 h after the morning intake of MMFdt. The concentrations of mycophenolic acid (MPA) in plasma were assayed with a validated reverse-phase high-performance liquid chromatography method. UGT1A8 518C > G, UGT1A9 -275T > A, UGT1A9 -2152C > T, UGT2B7 211G > T, SLC O 1B1 521T > C polymorphism were determined by Sanger sequencing. Nonlinear mixed effects modeling was used to establish the population pharmacokinetics (PPK) model. The predictability and stability of the model were internally evaluated by the goodness of fit plots, visual prediction check, normalized prediction errors, and bootstraps. Results: A two-compartment model with first-order absorption and first-order elimination was established with 115 MPA concentrations from 20 pediatric patients. The final model were: CL/F (L/h) = 14.8×(WT/7.5)0.75×(DOSE/11.16)0.452×е0.06, Ka (h−1) = 2.02×(WT/7.5)−0.25, Vc/F (L) = 6.01×(WT/7.5), Vp/F (L) = 269 (fixed), Q/F (L/h) = 15.4×(WT/7.5)0.75×е1.39. Where CL/F was the apparent clearance rate, Ka was the absorption rate constant, Vc/F was the apparent distribution volume of the central compartment, Vp/F was the apparent distribution volume of the peripheral compartment, Q/F was the atrioventricular clearance rate, WT was the body weight of the subject, and DOSE was the MMFdt administered dose. The model indicated there was large inter-individual variability in CL/F and Q/F after multiple dosing of MMFdt. Internal evaluation results showed that the final model had good stability and prediction performance. Conclusion: A stable and predictive population pharmacokinetic model of MMFdt in pediatric patients after the early stage of liver transplantation was established. The pediatric patient’s weight and the dose of MMFdt can be a reference to adjust the MMFdt dose.

Testing the Ability of Convolutional Neural Networks to Learn Radiomic Features

Background and objectiveRadiomics and deep learning have emerged as two distinct approaches to medical image analysis. However, their relative expressive power remains largely unknown. Theoretically, hand-crafted radiomic features represent a mere subset of features that neural networks can approximate, thus making deep learning a more powerful approach. On the other hand, automated learning of hand-crafted features may require a prohibitively large number of training samples. Here we directly test the ability of convolutional neural networks (CNNs) to learn and predict the intensity, shape, and texture properties of tumors as defined by standardized radiomic features. MethodsConventional 2D and 3D CNN architectures with an increasing number of convolutional layers were trained to predict the values of 16 standardized radiomic features from real and synthetic PET images of tumors, and tested. In addition, several ImageNet-pretrained advanced networks were tested. A total of 4000 images were used for training, 500 for validation, and 500 for testing. ResultsFeatures quantifying size and intensity were predicted with high accuracy, while shape irregularity and heterogeneity features had very high prediction errors and generalized poorly. For example, mean normalized prediction error of tumor diameter with a 5-layer CNN was 4.23 ± 0.25, while the error for tumor sphericity was 15.64 ± 0.93. We additionally found that learning shape features required an order of magnitude more samples compared to intensity and size features. ConclusionsOur findings imply that CNNs trained to perform various image-based clinical tasks may generally under-utilize the shape and texture information that is more easily captured by radiomics. We speculate that to improve the CNN performance, shape and texture features can be computed explicitly and added as auxiliary variables to the networks, or supplied as synthetic inputs.

Wireless Signal Propagation Prediction Based on Computer Vision Sensing Technology for Forestry Security Monitoring.

In this paper, Computer Vision (CV) sensing technology based on Convolutional Neural Network (CNN) is introduced to process topographic maps for predicting wireless signal propagation models, which are applied in the field of forestry security monitoring. In this way, the terrain-related radio propagation characteristic including diffraction loss and shadow fading correlation distance can be predicted or extracted accurately and efficiently. Two data sets are generated for the two prediction tasks, respectively, and are used to train the CNN. To enhance the efficiency for the CNN to predict diffraction losses, multiple output values for different locations on the map are obtained in parallel by the CNN to greatly boost the calculation speed. The proposed scheme achieved a good performance in terms of prediction accuracy and efficiency. For the diffraction loss prediction task, 50% of the normalized prediction error was less than 0.518%, and 95% of the normalized prediction error was less than 8.238%. For the correlation distance extraction task, 50% of the normalized prediction error was less than 1.747%, and 95% of the normalized prediction error was less than 6.423%. Moreover, diffraction losses at 100 positions were predicted simultaneously in one run of CNN under the settings in this paper, for which the processing time of one map is about 6.28 ms, and the average processing time of one location point can be as low as 62.8 us. This paper shows that our proposed CV sensing technology is more efficient in processing geographic information in the target area. Combining a convolutional neural network to realize the close coupling of a prediction model and geographic information, it improves the efficiency and accuracy of prediction.

Braid representation of the shallow water acoustic channel to interpret temporal evolution of multipath arrivals

We present a morphological representation of the underwater acoustic channel that employs geometric braids to gain real-time knowledge of the dominant multipath activity. The key idea is to incorporate braid manifolds in tandem with non-convex mixed norm optimization techniques to track the temporal evolution of multipath arrivals. Specifically, we focus on channel braids that manifest as rapidly fluctuating high-energy taps in the delay spread as well as the delay-Doppler scattering function. Multiple braids can also be topologically combined using braid operations to interpret oceanic phenomena such as caustics and surface wave focusing, among others. We present techniques for adaptively updating the channel braids and their overlap patterns to reflect the temporal evolution of the shallow water acoustic channel. We evaluate the performance of proposed morphological channel estimation technique in terms of the normalized prediction error and computational time. Results based on numerical channel simulations based on diverse oceanic and experimental conditions as well as experimental field data from the SPACE08 experiment will be presented.

Dynamical properties of LFPs from mice with unilateral injection of TeNT

Local field potential (LFP) recordings were performed from the visual cortex (V1) of a focal epilepsy mouse model. Epilepsy was induced by a unilateral injection of the synaptic blocker tetanus neurotoxin (TeNT). LFP signals were simultaneously recorded from V1 of both hemispheres of each animal under acute and chronic conditions (i.e. during and after the period of TeNT action). All data were analysed by using nonlinear time series methods. Suitable values of the lag time and embedding dimension for phase space reconstruction were estimated by employing well-known methods. The results showed that lag times are sensitive to the presence of TeNT. Interestingly, TeNT promoted an increase in the level of linear and nonlinear correlation of LFP signals. The values of the embedding dimension failed to show any dependence on the presence of the neurotoxin. However, a local nonlinear prediction method showed that the presence of TeNT increases the predictability, quantified by the normalized prediction error, of the neural recordings. From a neurophysiological point of view, the above results suggest that TeNT injected in one hemisphere strongly impacts the local electrical activity of the neural populations in the opposite hemisphere. We hypothesize that this could arise from a qualitative and quantitative alteration of the transmission properties of the callosal fibers.

Adaptive Neuro-Fuzzy Inference System Models for Force Prediction of a Mechatronic Flexible Structure

This paper presents the results obtained from a research work investigating the performance of different Adaptive Neuro-Fuzzy Inference System (ANFIS) models developed to predict excitation forces on a dynamically loaded flexible structure. For this purpose, a flexible structure is equipped with acceleration transducers at each degree of freedom and a force transducer for validation and training. Two types of models were developed; the first type uses current accelerations only while the second type considers both the current accelerations and the historical values. The models are trained using data obtained from applying a random excitation force on the flexible structure. The performance of the developed models is evaluated by analyzing the prediction capabilities based on a normalized prediction error. The frequency domain is considered to analyze the similarity of the frequencies in the predicted and the original force signal. For a selection of the best models, a more advanced performance analysis is carried out. This includes application of the trained models to deterministic and non- deterministic excitation forces with different excitation frequencies and amplitudes. Additionally, the influence of the sampling frequency and sensor location on the model performance is investigated. The results obtained in this paper show that ANFIS models can be used to set up reliable force predictors for dynamical loaded flexible structures, when a certain degree of inaccuracy is accepted. Furthermore, the comparison study points out that the transducer location is crucial for the model performance. However, there exists no general solution for the final selection of models. The findings showed that the model type employing historical values gives better predictions when operating in their trained regions while the models using only current values have generally higher prediction errors in trained regions but are less sensitive to changes of the system dynamics history.

Model selection for long-memory processes in the spectral domain

In this paper, we study a novel empirical criterion for identifying the type of a long-memory time series. The proposed rule for selecting an underlying model suitable to the data is based on the comparison between the normalized prediction errors of the generalized Whittle estimator (a parametric spectral estimator) applied to two or more candidate models. We test this approach by two applications of the procedure: for comparing two distinct statistical models to adjust the data, and for assessing the significance of increasing the number of parameters within a given class of models. Due to the heuristic nature of the method, we test the statistic numerically for several classes of stochastic processes, namely Gaussian processes with long-range dependence (LRD), M/G/∞ (non-Gaussian) processes with LRD, non-stationary processes, and non-linear heteroscedastic models. The numerical results demonstrate that the proposed statistic exhibits good power, is robust and not computationally expensive.

Computation of the Normalized Prediction Error of the Electroencephalogram Signal

In this paper, the normalized prediction error of the electroencephalogram (EEG) signal recorded at five different mental tasks was computed. The results indicate that there exists predictability in the EEG signal beyond the baseline prediction of the mean and the one-step-ahead normalized prediction error of EEG signal vary greatly when different mental tasks are implemented, which implies that the one-step-ahead normalized prediction error can be employed as a feature of EEG signal to distinct different mental tasks. For different subjects, the one-step-ahead normalized prediction error vary greatly even the EEG signal are recorded from the same electrode under the same mental task, which implies that the subjects’ individual differences should be considered adequately when the one-step-ahead normalized prediction error is employed to distinct different mental tasks. DOI: http://dx.doi.org/10.11591/telkomnika.v10i5.1268

Shallow water channel tracking using gradient-based mixed norm optimization techniques.

The underwater acoustic channel is challenging to estimate and track due to channel dynamism, the ill-conditioned nature of the estimation problem, as well as the need to optimize over a complex field. An extension of previous work is presented where the sparse channel coefficients in the delay and delay-Doppler spread domains are estimated using mixed norm convex optimization. The estimation algorithm is extended to track the time-varying channel using a gradient descent approach, trading off optimality for tracking efficiency. A computationally inexpensive approximation of the dynamic gradient is used rather than re-computing the precise gradient every time. Results presented are based on field data collected over 15-m depth and a range of wave heights and distances using the normalized prediction error as the performance metric. Comparison of performance between tracking the delay-only and delay-Doppler coefficients is presented for different weightings on the mixed norm between L1 and L2 measures and over different lags. Performance and computational efficiency of the proposed tracking algorithm are compared to conventional sparse estimation techniques. [Work supported in part by ONR Grants N00014-05-10085 and N00014-07-10184 and in part by Postdoctoral Scholar Program at the Woods Hole Oceanographic Institution, with funding provided by the Doherty Foundation.]

Constraints on shallow low-viscosity zones in Northern Europe from future GOCE gravity data

SUMMARY Current constraints on the process of glacial-isostatic adjustment in Northern Europe are mainly provided by relative sea level data and GPS measurements. Due to a lack of resolving power in the shallow Earth (down to ∼200 km), these data sets only provide weak constraints on the shallow viscosity structure and the thickness of the lithosphere. Future high-resolution gravity data, as expected from ESA's Gravity field and steady-state Ocean Circulation Explorer (GOCE) to be launched 2009 March 16, are predicted to provide additional information on the shallow Earth, especially the viscosity structure. However, mass inhomogeneities due to chemical and thermal anomalies are expected to interfere with the gravity signals induced by shallow low-viscosity structures. We test therefore if heatflow data and laboratory-derived creep laws for the crust (plagioclase feldspars) and shallow upper mantle (olivine) can provide additional information on the shallow Earth. We show estimates of lithospheric thickness and viscosity that can be expected in the shallow Earth. Using a mechanical model based on the commercially available finite-element package ABAQUS and representative creep laws, we generate predictions of deformation-induced geoid height variations for Northern Europe. There, lateral heterogeneities in the shallow Earth are induced based on heatflow data. We use the RSES ice-load history to force our mechanical model, and we test the sensitivity of our predictions using the ICE-5G ice-load history. We show that perturbations, that is, differences to a background model, due to shallow low-viscosity structures are one to two orders of magnitude larger than the predicted accuracy of GOCE, which is at the cm-level for a resolution of about 100 km. Moreover, some features in geoid height perturbations are robust to changes in composition and creep regime, and have therefore a spatial signature that is representative for low-viscosity structures, without detailed a priori knowledge on these structures. We argue that these signatures are therefore more likely to be detectable by GOCE. Finally, we show, using normalized prediction errors, that GOCE is sensitive to the creep regime in the lower crust but not to the composition, at least not for the plagioclase feldspars used here. These conclusions are, in general, independent of assumptions (creep regime in the shallow upper mantle, ice-load history) on the background model. However, if the wrong background model is assumed, we can no longer predict the correct properties of the lower crust, because prediction errors are larger than 60 per cent.

Neural Network Model of Plasma Charging Damage on MOSFET Device

Plasma-induced charging damage can degrade device reliability considerably. Constructing charging damage model is important since it can be used for qualitative parameter effect analysis as well as for process optimization. In this study, a backpropagation neural network was used to capture causal relationships between process parameters and plasma charging-damaged metal-oxide semiconductor (MOS) device. For a systematic modeling, plasma charging damage process was characterized by means of a face-centered Box Wilson experiment. As a charge collector, aluminum antennae were connected to MOS device. The prediction performance of BPNN was optimized as a function of training factors. Electrical device characteristics modeled include threshold voltage (V), subthreshold swing (S), and transconductance (G). The normalized prediction errors are 1.84, 2.99, and 1.87{%} for V, S, and G, respectively. The optimized models were compared to statistical regression and fuzzy logic models. The comparison revealed that neural network models yielded more improved predictions over either of them. The improvement was even more than 40{%} for V and S.

A test method for weak nonlinearity in time series

Nonlinearity is necessary for time series to be treated as chaotic time series. A new test statistic for nonlinearity, which is based on the ratio of the multistep normalized prediction error with respect to linear AR models and nonlinear AR models, is used to detect the weak nonlinear components contained in time series by the surrogate data method. Taking example for Lorenz time series, the effect of related parameters for test statistic estimation is analyzed. By the nonlinearity tests for chaotic time series, the proposed test statistic δNAR has better discrimination power for weak nonlinearity than the test statistic δAIC based on AIC rules and the zeroth order nonlinear prediction error δZP, which shows that the proposed test statistic has strong adaptive abilities for time series. And, for different time series, the parameters with best nonlinearity discrimination performance are kept constant. The stabilization of parameters facilitates the nonlinearity test for other time series.

A Data mining framework of noninvasive intracranial pressure assessment

The only established technique for intracranial pressure (ICP) measurement is an invasive direct procedure that has been accepted into routine usage in neurosurgical services. However, there are many other scenarios where a noninvasive assessment of ICP is highly desirable. To make a full use of the vast amount of signals collected in a typical neurosurgical service environment to realize such a noninvasive procedure of ICP assessment, a general data mining framework is proposed in the present work. As a particular implementation of the proposed framework where continuous arterial blood pressure (ABP) and cerebral blood flow velocity (CBFV) are available, we propose to simulate the unobserved ICP as the output of a model built from a database composed of arterial blood pressure, cerebral blood flow velocity and ICP. This model was discovered by exploring the database with the hemodynamic features extracted from measured ABP and CBFV. This approach was evaluated using a database composed of 30-min long measurements from nine traumatic brain injury patients. The approach achieved significant improvements of ICP simulation accuracy over several existing noninvasive ICP assessment methods. Particularly, its median normalized prediction error for ICP is 39% compared to 51% as obtained by an existing method and its median correlation coefficient between estimated and measured normalized ICP is 0.80 compared to 0.35 achieved by the existing method. The proposed framework is flexible in incorporating other relevant signals besides ABP, CBFV and ICP into the database. It allows for designing new hemodynamic feature vectors and for adopting new models for ICP estimation. Hence it is worthwhile to evaluate the method using a larger database and further develop the framework.

Signal processing procedure for non-destructive test of concrete structure integrity

A signal processing procedure is proposed for testing concrete integrity using impact by shock wave and Laser Doppler velocimetry. The concrete surface vibration signal is first segmented into successive frames. The signal in each frame is analyzed using a linear prediction model to calculate the normalized prediction error. This error indicates the degree to which the signal is modeled in the form of an exponentially decaying sinusoid, and the presence of sub-surface flaws in the concrete can be determined from its value. When a defect in the concrete is detected, the frequency of the decaying sinusoid is calculated to estimate the depth and size of the crack. The proposed procedure was applied to computer-generated signals and actual vibration signals produced by shock waves, and the success of the procedure was verified.

How Accurate Channel Prediction Needs to be for Transmit-Beamforming With Adaptive Modulation Over Rayleigh MIMO Channels?

Adaptive modulation improves the system throughput considerably by matching transmitter parameters to time-varying wireless fading channels. Crucial to adaptive modulation is the quality of channel state information at the transmitter. In this paper, we first present a channel predictor based on pilot symbol assisted modulation for multiple-input multiple-output Rayleigh fading channels. We then analyze the impact of the channel prediction error on the bit error rate performance of a transmit-beamformer with adaptive modulation that treats the predicted channels as perfect. Our numerical results reveal the critical value of the normalized prediction error, below which the predicted channels can be treated as perfect by the adaptive modulator; otherwise, explicit consideration of the channel imperfection must be accounted for at the transmitter.

FOR NONLINEARITY IN THE HEARTBEAT TIME SERIES

Hearbeat time series obtained by Holter monitoring of 10 healthy subjects are studied using a nonlinear predictor S map . The normalized prediction error was calculated as a function of the model control parameter in order to get information on the amount of nonlinearity in the data. Moreover, to search for possible chaotic behavior, the linear correlation coefficient between predicted and real values was calculated as a function of the prediction time. The results of this analysis reveal no clear evidence of chaoticness or nonlinearity in the data. Moreover, for 8 subjects out of 10, the predictability during sleep is better than during the daytime.

Stabilized Least-Squares Estimators for Time-Variant Processes

Abstract The paper presents the analysis of a class of modified recursive least-squares algorithms (RLS-SF) devoted to track parameters of time-varying plants in the presence of poor excitation. The algorithms are stabilized in the sense of bounding the covariance matrix from above and below and thus avoiding blow-up. They combine linear-forgetting recursive least-squares algorithm with exponential covariance matrix stabilization. The numerical burden imposed by the modification is of O(n) and thus can be neglected. Under the assumption that the data are generated by a deterministic linear time-invariant system the RLS-SI algorithm is exponentially convergent for persistently exciting signals. For not persistently exciting signals the normalized prediction errors and estimate changes are square summable and the estimates are bounded. Asymptotic behavior of the covariance matrix is considered for not persistently exciting signals: the eigenvectors of the matrix can be asymptotically partitioned into two groups, where the first group spans entirely the excited subspace and the second one is orthogonal to the excited subspace. The good perfonnance of the algorithm is verifyed via simulation studies.

Direction finding in the presence of colored noise by candidate identification

A method is developed for identifying correct angles of arrivals from a set of candidate angles that contains spurious angles and the true ones, under the assumption that the colored noise is an autoregressive process and that a uniform linear array of sensors is used. The procedure is based on a relation derived for the higher-order reflection coefficients corresponding to the autocorrelation sequence of the signal plus noise. It is shown that the higher-order reflection coefficients of a set of plane waves impinging on a uniform linear array of sensors, in the presence of an unknown spatially autoregressive colored noise process, are equal to an order independent linear combination of the values of the Fourier transform of the corresponding optimal linear least squares normalized prediction error filter at the electric phase angles of the plane waves. The magnitudes of the coefficients are then used to decide whether a given candidate direction of arrival corresponds to an actual plane wave or not. The estimation of candidate angles is also briefly discussed.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Replacing tracheoesophageal voicing sources using LPC synthesis

The feasibility of using the linear predictive coding (LPC) technique to replace the voicing sources of tracheoesophageal speech was explored. Four vowels, [i], [a], [e], [u], and one diphthong [ou], produced by two male and two female tracheoesophageal speakers were analyzed by the LPC autocorrelation method. Normalized prediction error functions were used to choose the algorithm and the control parameters of the analysis. Poles of the vocal tract transfer function were selected from frames whose normalized prediction errors were close to minimum with criteria derived from transfer functions of normally produced vowels. Vowels were synthesized with the reconstructed transfer function and a synthesized excitation input. Results of an identification task indicated that the synthesized vowels were highly intelligible, and the gender of the speaker was better identified from the synthesized vowels than from the original tracheoesophageal vowels.