Attention and prediction error as mechanisms for theory protection?

Wind power prediction of wind farm plays a decisive role in stable electric power system operation.The BP neural network’s basic principle was introduced, and the numerical weather prediction (NWP) data and power data of wind farm as the training data of BP neural network was selected and trained; a linear regression model about the sample prediction error was presented, which considers the coupling relationship between the individual sample prediction error, the individual sample prediction error of BP neural network was selected as the regression factor, the individual sample prediction result of BP neural network was modified. As the modified prediction results performing, the prediction algorithm of short-term wind power considering the sample prediction error correction, has good self-learning and adaptive ability of BP neural network. It has overcome the shortcoming that the BP neural network has only considered the overall the prediction error of training samples, but without considered the prediction error of individual samples. This has further improved the prediction accuracy of BP neural network.

Mechanisms Underlying Dopamine-Mediated Reward Bias in Compulsive Behaviors

Information received from the visual and chemical senses is qualitatively different. For prey species in aquatic environments, visual cues are spatially and temporally reliable but risky as the prey and predator must often be in close proximity. Chemical cues, by contrast, can be distorted by currents or linger and thus provide less reliable spatial and temporal information, but can be detected from a safe distance. Chemical cues are therefore often the first detected and may provide a context in which prey respond to subsequent ambiguous cues (“context hypothesis”). Depending on this context, early chemical cues may also alert prey to attend to imminent cues in other sensory modalities (“alerting hypothesis”). In the context of predation risk, for example, it is intuitive that individuals become more responsive to subsequent ambiguous cues across sensory modalities. Consistent with the context hypothesis, guppies, Poecilia reticulata, exposed to conspecific alarm cue reduced activity, a classic fright response among fish, in response to a water disturbance more than those exposed to cues of unharmed conspecifics or a water control. Despite this reduction in activity, guppies exposed to alarm cue were more attentive to visual cues than those exposed to the other chemical cues, as predicted by the alerting hypothesis. These responses contrasted with those of guppies exposed to chemical cues of undisturbed, unharmed conspecifics, which were relatively unaffected by the disturbance. This is the first study indicating that unambiguous cues detected by one sensory modality affect animal responses to subsequent ambiguous multimodal cues.Significance statementIn moving water, chemical cues can be detected over longer distances than visual cues; they may therefore be detected first and alert animals to imminent visual cues. This effect is likely to be particularly important if these chemical cues are indicative of predation. I investigated how different chemical cues affect (1) guppy response to an ambiguous water disturbance and (2) their responsiveness to subsequent ambiguous visual cues. Guppies based their responses to ambiguous cues on the context implied by chemical cues: those exposed to chemical cues indicative of predation reduced activity, a classic fright response, but increased responsiveness to visual cues, relative to those exposed to control chemical cues. This is the first study to show that unambiguous cues detected by one sense affect animal responses to ambiguous cues detected by other senses.Electronic supplementary materialThe online version of this article (doi:10.1007/s00265-016-2076-4) contains supplementary material, which is available to authorized users.

Predicting postoperative intraocular lens position and refraction

Substance-dependent individuals make poor decisions on the Iowa Gambling Task, a reward-related decision-making task that involves risk and uncertainty. Task performance depends on several factors, including how sensitive individuals are to feedback and how well they learn based on such feedback. A physiological signal that guides decision making based on feedback is prediction error. The authors investigated whether disruptions in the neural systems underlying prediction error processing in substance-dependent individuals could account for decision-making performance on a modified Iowa Gambling Task. Thirty-two substance-dependent individuals and 30 healthy comparison subjects played a modified version of the Iowa Gambling Task during MR scanning. Trial-to-trial behavior and functional MRI (fMRI) blood-oxygen-level-dependent (BOLD) signal were analyzed using a computational model of prediction error based on internal expectancies. The authors investigated how well BOLD signal tracked prediction error in the striatum and the orbitofrontal cortex as well as over the whole brain in patients relative to comparison subjects. Compared with healthy subjects, substance-dependent patients were less sensitive to loss compared with gain, made less consistent choices, and performed worse on the modified Iowa Gambling Task. The ventral striatum and medial orbitofrontal cortex did not track prediction error as strongly in patients as in healthy subjects. Weaker tracking of prediction error in substance-dependent relative to healthy individuals suggests that altered frontal-striatal error learning signals may underlie decision-making impairments in drug abusers. Computational fMRI may help bridge the knowledge gap between physiology and behavior to inform research aimed at substance abuse treatment.

SummaryHow the brain uses success and failure to optimize future decisions is a long-standing question in neuroscience. One computational solution involves updating the values of context-action associations in proportion to a reward prediction error. Previous evidence suggests that such computations are expressed in the striatum and, as they are cognitively impenetrable, represent an unconscious learning mechanism. Here, we formally test this by studying instrumental conditioning in a situation where we masked contextual cues, such that they were not consciously perceived. Behavioral data showed that subjects nonetheless developed a significant propensity to choose cues associated with monetary rewards relative to punishments. Functional neuroimaging revealed that during conditioning cue values and prediction errors, generated from a computational model, both correlated with activity in ventral striatum. We conclude that, even without conscious processing of contextual cues, our brain can learn their reward value and use them to provide a bias on decision making.

Rifampicin exhibits complexities in its pharmacokinetics (PK), including high inter-occasion variability (IOV), which is challenging for dose individualization. Model-informed precision dosing (MIPD) can be used to optimize individual doses. In this simulation-based study we investigated the magnitude of IOV in rifampicin PK on an exposure level, the impact of not acknowledging IOV when performing MIPD, and the number of sampling occasions needed to forecast the dose. Subjects with drug-susceptible tuberculosis (TB) were simulated from a previously developed population PK model. To explore the magnitude of IOV, the area under the plasma concentration-time curve from time zero up to 24 h (AUC0–24h) after 35 mg/kg in the typical individual was simulated for 1,000 sampling occasions at steady-state. The impact of ignoring IOV for dose predictions was investigated by comparing the prediction error of a MIPD approach including IOV to an approach ignoring IOV. Furthermore, the number of sampling occasions needed to predict individual doses using a MIPD approach was assessed. The AUC0–24h in the typical individual varied substantially between simulated sampling occasions [95% prediction interval (PI): 122.2 to 331.2 h mg/L], equivalent to an IOV in AUC0–24h of 25.8%, compared to an inter-individual variability of 25.4%. The median of the individual prediction errors using a MIPD approach incorporating IOV was 0% (75% PI: −14.6% to 0.0%), and the PI for the individual prediction errors was narrower with than without IOV (median: 0%, 75% PI: −14.6% to 20.0%). The most common target dose in this population was forecasted correctly in 95% of the subjects when IOV was included in MIPD. In subjects where doses were not predicted optimally, a lower dose was predicted compared to the target, which is favorable from a safety perspective. Moreover, the imprecision (relative root mean square error) and bias in predicted doses using MIPD with IOV decreased statistically significant when a second sampling occasion was added (difference in imprecision: −9.1%, bias: −7.7%), but only marginally including a third (difference in imprecision: −0.1%, bias: −0.1%). In conclusion, a large variability in exposure of rifampicin between occasions was shown. In order to forecast the individual dose correctly, IOV must be acknowledged which can be achieved using a MIPD approach with PK information from at least two sampling occasions.

Radial artery tonometry: moderately accurate but unpredictable technique of continuous non-invasive arterial pressure measurement

The inferential estimation of a polymer melt index in an industrial polymerization process using aggregated neural networks is presented in this paper. The difficult‐to‐measure polymer melt index is estimated from easy‐to‐measure process variables, and their relationship is estimated using aggregated neural networks. The individual networks are trained on bootstrap re‐samples of the original training data by a sequential training algorithm. In this training method, individual networks, within a bootstrap aggregated neural network model, are trained sequentially. The first network is trained to minimize its prediction error on the training data. In the training of subsequent networks, the training objective is not only to minimize the individual networks' prediction errors but also to minimize the correlation among the individual networks. Training is terminated when the aggregated network prediction performance on the training and testing data cannot be further improved. Application to real industrial data demonstrates that the polymer melt index can be successfully estimated using an aggregated neural network.

A sequential training method for developing bootstrap aggregated neural network models is proposed in this paper. In this method, individual networks within a bootstrap aggregated neural network model are trained sequentially. The first network is trained to minimise its prediction error on the training data. In the training of subsequent networks, the training objective is not only to minimise the individual networks' prediction errors but also to minimise the correlation among the individual networks. Training data sets for the individual networks are different and are generated through bootstrap re-sampling of the original training data set. Training is terminated when the aggregated network prediction performance cannot be further improved. An application example demonstrates the superior performance of this neural network training strategy.

Disproportionate reactions to unexpected stimuli in the environment are a cardinal symptom of posttraumatic stress disorder (PTSD). Here, we test whether these heightened responses are associated with disruptions in distinct components of reinforcement learning. Specifically, using functional neuroimaging, a loss-learning task, and a computational model-based approach, we assessed the mechanistic hypothesis that overreactions to stimuli in PTSD arise from anomalous gating of attention during learning (i.e., associability). Behavioral choices of combat-deployed veterans with and without PTSD were fit to a reinforcement learning model, generating trial-by-trial prediction errors (signaling unexpected outcomes) and associability values (signaling attention allocation to the unexpected outcomes). Neural substrates of associability value and behavioral parameter estimates of associability updating, but not prediction error, increased with PTSD during loss learning. Moreover, the interaction of PTSD severity with neural markers of associability value predicted behavioral choices. These results indicate that increased attention-based learning may underlie aspects of PTSD and suggest potential neuromechanistic treatment targets.

In this study, a series of tests have been conducted in order to investigate the machinability evaluation of austenitic Hadfield manganese steel in the Wire Electrical Discharge Machine (WEDM). Experimental investigations have been carried out to relate the effect of input machining parameters such as pulse on-time (Ton), pulse off-time (Toff), wire feed (WF), and average gap voltage (V) on the wire offset in WEDM. No analytical approach gives the exact amount of offset required in WEDM and hence experimental study has been undertaken. In this paper, a mathematical model has been developed to model the machinability evaluation through the response surface methodology (RSM) capable of predicting the response parameter as a function of Ton, Toff, WF and V. The samples are tested and their average prediction error has been calculated taking the average of all the individual prediction errors. The result shows that this mathematical model reflects the independent, quadratic and interactive effects of the various machining parameters on cutting speed in WEDM process.

In machine learning and its risk-sensitive applications (e.g. medicine, engineering, business), the reliability estimates for individual predictions provide more information about the individual prediction error (the difference between the true label and regression prediction) than the average accuracy of predictive model (e.g. relative mean squared error). Furthermore, they enable the users to distinguish between more and less reliable predictions. The empirical evaluations of the existing individual reliability estimates revealed that the successful estimates’ performance depends on the used regression model and on the particular problem domain. In the current paper, we focus on that problem as such and propose and empirically evaluate two approaches for automatic selection of the most appropriate estimate for a given domain and regression model: the internal cross-validation approach and the meta-learning approach. The testing results of both approaches demonstrated an advantage in the performance of dynamically chosen reliability estimates to the performance of the individual reliability estimates. The best results were achieved using the internal cross-validation procedure, where reliability estimates significantly positively correlated with the prediction error in 73% of experiments. In addition, the preliminary testing of the proposed methodology on a medical domain demonstrated the potential for its usage in practice.

Physiologically based pharmacokinetic (PBPK) modeling is a mechanistic concept, which helps to judge the effects of biopharmceutical properties of drug product such as in vitro dissolution on its pharmacokinetic and in vivo performance. With the application of virtual bioequivalence (VBE) study, the drug product development using model-based approach can help in evaluating the possibility of extending BCS-based biowaiver. Therefore, the current study was intended to develop PBPK model as well as in vitro in vivo extrapolation (IVIVE) for BCS class III drug i.e. cefadroxil.A PBPK model was created in GastroPlus™ 9.8.3 utilizing clinical data of immediate-release cefadroxil formulations. By the examination of simulated and observed plasma drug concentration profiles, the predictability of the proposed model was assessed for the prediction errors. Furthermore, mechanistic deconvolution was used to create IVIVE, and the plasma drug concentration profiles and pharmacokinetic parameters were predicted for different virtual formulations with variable cefadroxil in vitro release. Virtual bioequivalence study was also executed to assess the bioequivalence of the generic verses the reference drug product (Duricef®).The developed PBPK model satisfactorily predicted Cmax and AUC0-t after cefadroxil single and multiple oral dose administrations, with all individual prediction errors within the limits except in a few cases. Second order polynomial correlation function obtained accurately predict in vivo drug release and plasma concentration profile of cefadroxil test and reference (Duricef®) formulation. The VBE study also proved test formulation bioequivalent to reference formulation and the statistical analysis on pharmacokinetic parameters reported 90% confidence interval for Cmax and AUC0-t in the FDA acceptable limits.The analysis found that a validated and verified PBPK model with a mechanistic background is as a suitable approach to accelerate generic drug development.

This study examined the ability of the HRindex model to accurately predict maximal oxygen uptake (O2max) across a variety of incremental exercise protocols. Ten men completed five incremental protocols to volitional exhaustion. Protocols included three treadmill (Bruce, UCLA running, Wellness Fitness Initiative [WFI]), one cycle, and one field (shuttle) test. The HRindex prediction equation (METs = 6 × HRindex − 5, where HRindex = HRmax/HRrest) was used to generate estimates of energy expenditure, which were converted to body mass-specific estimates of O2max. Estimated O2max was compared with measured O2max. Across all protocols, the HRindex model significantly underestimated O2max by 5.1 mL·kg−1·min−1 (95% CI: −7.4, −2.7) and the standard error of the estimate (SEE) was 6.7 mL·kg−1·min−1. Accuracy of the model was protocol-dependent, with O2max significantly underestimated for the Bruce and WFI protocols but not the UCLA, Cycle, or Shuttle protocols. Although no significant differences in O2max estimates were identified for these three protocols, predictive accuracy among them was not high, with root mean squared errors and SEEs ranging from 7.6 to 10.3 mL·kg−1·min−1 and from 4.5 to 8.0 mL·kg−1·min−1, respectively. Correlations between measured and predicted O2max were between 0.27 and 0.53. Individual prediction errors indicated that prediction accuracy varied considerably within protocols and among participants. In conclusion, across various protocols the HRindex model significantly underestimated O2max in a group of aerobically fit young men. Estimates generated using the model did not differ from measured O2max for three of the five protocols studied; nevertheless, some individual prediction errors were large. The lack of precision among estimates may limit the utility of the HRindex model; however, further investigation to establish the model's predictive accuracy is warranted.