Propagation Of Prediction Errors Research Articles

Unsupervised Characterization of Prediction Error Markers in Unisensory and Multisensory Streams Reveal the Spatiotemporal Hierarchy of Cortical Information Processing.

Elicited upon violation of regularity in stimulus presentation, mismatch negativity (MMN) reflects the brain's ability to perform automatic comparisons between consecutive stimuli and provides an electrophysiological index of sensory error detection whereas P300 is associated with cognitive processes such as updating of the working memory. To date, there has been extensive research on the roles of MMN and P300 individually, because of their potential to be used as clinical markers of consciousness and attention, respectively. Here, we intend to explore with an unsupervised and rigorous source estimation approach, the underlying cortical generators of MMN and P300, in the context of prediction error propagation along the hierarchies of brain information processing in healthy human participants. The existing methods of characterizing the two ERPs involve only approximate estimations of their amplitudes and latencies based on specific sensors of interest. Our objective is twofold: first, we introduce a novel data-driven unsupervised approach to compute latencies and amplitude of ERP components accurately on an individual-subject basis and reconfirm earlier findings. Second, we demonstrate that in multisensory environments, MMN generators seem to reflect a significant overlap of "modality-specific" and "modality-independent" information processing while P300 generators mark a shift toward completely "modality-independent" processing. Advancing earlier understanding that multisensory contexts speed up early sensory processing, our study reveals that temporal facilitation extends to even the later components of prediction error processing, using EEG experiments. Such knowledge can be of value to clinical research for characterizing the key developmental stages of lifespan aging, schizophrenia, and depression.

Multi-task-based spatiotemporal generative inference network: A novel framework for predicting the highway traffic speed

Accurately predicting the highway traffic speed can reduce traffic accidents and transit time, and it also provides valuable reference data for traffic control in advance. Three essential elements should be considered in highway traffic speed prediction: (1) dynamic spatiotemporal correlation, (2) prediction error propagation elimination, and (3) traffic speed heterogeneity on the highway network. A multi-task-based spatiotemporal generative inference network (MT-STGIN) is proposed to address the above challenges. 11https://doi.org/10.24433/CO.8500259.v1. MT-STGIN consists of three parts: an encoder based on spatiotemporal correlation extraction block (ST-Block), a decoder based on masked ST-Block and generative inference, and multi-task learning. The encoder is first used to extract the dynamic spatiotemporal correlation of the highway network. The decoder concentrates on correlating historical and - target sequences and generates the target hidden outputs rather than a dynamic step-by-step decoding way. Finally, a multi-task learning method is used to predict traffic speed on different types of road segments because of heterogeneity and shares the underlying network parameters. The evaluation experiments use the monitoring data of the highway in Yinchuan City, Ningxia Province, China. The experimental results demonstrate that the performance of the proposed prediction model is better than that of the baselines, and it can efficiently solve the problem of long-term highway speed prediction.

Same but different: The latency of a shared expectation signal interacts with stimulus attributes

Predictive coding theories assert that perceptual inference is a hierarchical process of belief updating, wherein the onset of unexpected sensory data causes so-called prediction error responses that calibrate erroneous inferences. Given the functionally specialised organisation of visual cortex, it is assumed that prediction error propagation interacts with the specific visual attribute violating an expectation. We sought to test this within the temporal domain by applying time-resolved decoding methods to electroencephalography (EEG) data evoked by contextual trajectory violations of either brightness, size, or orientation within a bound stimulus. We found that following ∼170 ms post stimulus onset, responses to both size violations and orientation violations were decodable from physically identical control trials in which no attributes were violated. These two violation types were then directly compared, with attribute-specific signalling being decoded from 265 ms. Temporal generalisation suggested that this dissociation was driven by latency shifts in shared expectation signalling between the two conditions. Using a novel temporal bias method, we then found that this shared signalling occurred earlier for size violations than orientation violations. To our knowledge, we are among the first to decode expectation violations in humans using EEG and have demonstrated a temporal dissociation in attribute-specific expectancy violations.

A novel spatio-temporal generative inference network for predicting the long-term highway traffic speed

Accurately predicting the highway traffic speed can reduce traffic accidents and transit time, which is of great significance to highway management. Three essential elements should be considered in the long-term highway traffic speed prediction: (1) adaptability to speed fluctuation, (2) exploring the spatio-temporal correlation effectively, and (3) prediction of non-error propagation. This paper proposes a novel spatio-temporal generative inference network (STGIN) driven by data and long-term prediction. STGIN consists of three parts; semantic enhancement, spatio-temporal correlation extraction block (ST-Block), and generative inference. Semantic enhancement is first used to model the contextual semantics of traffic speed, improving the adaptability to speed fluctuations. The ST-Block is then used to extract the spatio-temporal correlations of the highway network. Finally, generative inference is used to pay attention to the correlation between historical- and target-sequences to generate the target hidden outputs rather than a dynamic step-by-step decoding way; it avoids long-term prediction error propagation in the spatial and temporal dimensions. The evaluation experiments use the monitoring data of highway in Yinchuan City, Ningxia Province, China. For long-term highway speed prediction, the experimental results demonstrate that the performance of the proposed method is better than that of the baseline methods.

Longitudinal evidence for attenuated local-global deviance detection as a precursor of working memory decline.

From the perspective of predictive coding, normal aging is accompanied by decreased weighting of sensory inputs and increased reliance on predictions, resulting in the attenuation of prediction errors in older age. Recent EEG research further revealed that the age-related shift from sensorium to predictions is hierarchy-selective, as older brains show little reduction in lower-level but significant suppression in higher-level prediction errors. Moreover, the disrupted propagation of prediction errors from the lower-level to the higher-level seems to be linked to deficient maintenance of information in working memory. However, it is unclear whether the hierarchical predictive processing continues to decline with advancing age as working memory. Here we longitudinally followed a sample of 78 participants from three age groups (including seniors, adults, and adolescents) over three years' time. Seniors exhibited largely preserved local processing (consisting of comparable MMN, delayed P3a, and comparable RON) but significantly compromised global processing (consisting of suppressed frontocentral negativity and suppressed P3b) in the auditory local-global paradigm. These electrophysiological responses did not change with the passing of time, unlike working memory which deteriorated with advancing age. Correlation analysis further showed that these electrophysiological responses signalling prediction errors are indicative of concurrent working memory. Moreover, there was a correlation between earlier predictive processing and later working memory but not between earlier working memory and later predictive processing. The temporal asymmetry suggested that the hierarchy-selective attenuation of prediction errors is likely a precursor of working memory decline.Significance StatementWhile predictive coding is postulated as a fundamental principle of brain function, little is known about its developmental trajectory in normal aging. This is in stark contrast with the large body of research on age-related decline in cognition such as working memory, which could be conceptualised as a derivative of the predictive coding mechanism. Here we provided longitudinal evidence that the hierarchy-selective attenuation of prediction errors manifests as a stable feature while working memory deteriorates with advancing age. Correlation analysis further suggested that the attenuation of prediction errors is likely a precursor of working memory decline, rendering it a potential predictor of working memory deterioration before any clinically relevant symptoms would manifest in the aging population.

Design of the high-dimensional SPKF algorithm and application

In the framework of Kalman filtering, the key points is the estimation of the covariance of prediction error which is used to calculate the Kalman filter gain. For the traditional algorithm, e.g. extended Kalman filter, the priori information is obtained on the basis of the Taylor expansion approximation of the nonlinear system. Because the linear propagation approximation of prior information cannot truly reflect the evolution of the probability distribution of the prediction errors in nonlinear systems, however, the traditional method affects the estimation accuracy of the extended Kalman filter. To solve these problems and be easy to apply to high-dimensional nonlinear systems, this paper improves the SPKF algorithm. Firstly, the prediction error propagation model is re-established to eliminate the approximation error of Taylor expansion in the derivation of the probability density function of the prediction error. Then, a Fisher matrix calculation method that is easier for high-dimensional system is obtained by defining the calculation method of Hilbert inner product of vector and matrix function. To illustrate the merits of the novel algorithm, this paper took the oxygen generation system by using electrolytic water in space station as a case study and conducted the state estimation by respectively using the extended Kalman filter, unscented Kalman filter and high-dimensional stochastic projection Kalman filter. The simulation results shown that the new algorithm can obtain higher estimation precision than the traditional algorithm.

Keeping the Breath in Mind: Respiration, Neural Oscillations, and the Free Energy Principle.

Scientific interest in the brain and body interactions has been surging in recent years. One fundamental yet underexplored aspect of brain and body interactions is the link between the respiratory and the nervous systems. In this article, we give an overview of the emerging literature on how respiration modulates neural, cognitive and emotional processes. Moreover, we present a perspective linking respiration to the free-energy principle. We frame volitional modulation of the breath as an active inference mechanism in which sensory evidence is recontextualized to alter interoceptive models. We further propose that respiration-entrained gamma oscillations may reflect the propagation of prediction errors from the sensory level up to cortical regions in order to alter higher level predictions. Accordingly, controlled breathing emerges as an easily accessible tool for emotional, cognitive, and physiological regulation.

A novel method for VR sickness reduction based on dynamic field of view processing

In this paper, we proposed a novel method for virtual reality (VR) sickness reduction based on dynamic field of view (FOV) processing. Dynamic FOV processing is performed based on the estimated VR sickness for each video frame. The level of sickness is estimated using VR sickness model, which is obtained by defining the relationship between the motion information and the measured VR sickness. For motion information analysis, subregion-based correspondence points tracking is used to efficiently remove outliers and prevent prediction error propagation. Amount of head dispersion is used as a quantitative VR sickness measure, which can be calculated from inertial measurement unit sensor in VR devices. The optimal FOV range was determined by experimentally validating a minimum FOV that can effectively reduce VR sickness with almost negligible loss in presence. The simulation results show a significant decrease of 37% compared to full FOV viewing, when FOV is dynamically varied between full and 60°.

Language ERPs reflect learning through prediction error propagation

Event-related potentials (ERPs) provide a window into how the brain is processing language. Here, we propose a theory that argues that ERPs such as the N400 and P600 arise as side effects of an error-based learning mechanism that explains linguistic adaptation and language learning. We instantiated this theory in a connectionist model that can simulate data from three studies on the N400 (amplitude modulation by expectancy, contextual constraint, and sentence position), five studies on the P600 (agreement, tense, word category, subcategorization and garden-path sentences), and a study on the semantic P600 in role reversal anomalies. Since ERPs are learning signals, this account explains adaptation of ERP amplitude to within-experiment frequency manipulations and the way ERP effects are shaped by word predictability in earlier sentences. Moreover, it predicts that ERPs can change over language development. The model provides an account of the sensitivity of ERPs to expectation mismatch, the relative timing of the N400 and P600, the semantic nature of the N400, the syntactic nature of the P600, and the fact that ERPs can change with experience. This approach suggests that comprehension ERPs are related to sentence production and language acquisition mechanisms.

Speed prediction from mobile sensors using cellular phone‐based traffic data

The formulation of data-driven short-term traffic state prediction models is highly dependent on the characteristics of collected data. Mobile sensors, specifically, on-board cellular phones (CPs) have proven success in wide scale real-time traffic data collection, in areas with limited traffic surveillance infrastructure. In this research, four short-term travel speed prediction models have been examined to cater the CP-based traffic data environment. Time-series concepts were adopted for speed prediction by autoregressive integrated moving average model and non-linear autoregressive exogenous model that is trained by neural networks. Alternatively, Bayesian networks (BNTs) and dynamic BNTs (DBNs) speed prediction models, from the graphical-based arena, have been investigated. The developed prediction models were tested in MATLAB environment on data from a simulation platform for 26-of-July corridor in Greater Cairo, Egypt. Testing results revealed the advantage of graphical-based models in restricting the propagation of prediction errors from one time step to the next. BNT reported a mean absolute percentage error (MAPE) of 6.31 ± 1.03, whereas the DBN model reported a MAPE of 5.34 ± 1.90.

How accurately can soil classes be allocated based on spectrally predicted physio-chemical properties?

Soil class maps are useful representations of the landscape distribution of holistic soil functions. However these are often only available as generalized classes at small cartographic scales. One reason is that allocating a soil profile to a class in most current soil classification system requires laboratory determination of many diagnostic soil properties. The advantage of reflectance spectroscopy along with the development of spectral libraries can provide a relatively low-cost solution to this problem. Reflectance spectroscopy has demonstrated its ability to rapidly predict soil physio-chemical properties; however prediction accuracy varies among soil properties. When properties predicted with different accuracies are used to substitute for traditional laboratory determinations in allocating a soil profile to a class, the resulting reliability of the allocation is questionable. The objective of this research is to explore whether the soil properties predicted by reflectance spectroscopy can be used to correctly allocate soil profiles into soil taxa at different hierarchical levels. Two hundred and six soil profiles were allocated to eight Orders, 12 Suborders, 23 Groups and 49 Subgroups according to Chinese Soil Taxonomy, with the help of ten soil properties predicted by spectra using ten-fold cross-validated PLSR modelling. The overall allocation accuracy at Order, Suborder, Group and Subgroup level was 98.5%, 98.5%, 87.7% and 76.0% respectively. These results show that soil reflectance spectroscopy can assist in allocation of profiles. When predicted soil properties with varying accuracy are used for soil allocation, propagation of prediction errors and model uncertainties must be considered. We propose the use of multiple indicators (RPD, confidence intervals, comparison of RMSE and threshold requirements) to evaluate the allocation results.

Data-driven modeling of milling and sieving operations in a wheat milling process

Industrial wheat milling processes are based on a gradual size reduction approach, with repeated milling and sieving steps to achieve appropriate breakage of the wheat kernel and its effective reduction into flour. This approach greatly increases the flour yield, but makes the process more difficult to understand and operate. In this respect, mathematical modeling can be very useful to optimize industrial milling operations by relating the operating variables to the final product quality and quantity.As a first step toward the modeling of an entire wheat milling process, in this study experimental data were generated from roller mills and plansifters, and the data were used to build multivariate statistical models enabling one to improve process understanding and to predict the particle size distribution of the milled material and of the sieved material, as well as the amount of material remaining on each sieve, from the known operating conditions and wheat moisture. The unit operation models were then concatenated, and the propagation of prediction errors was analyzed. The results are very satisfactory, in that a comprehension of the breakage mechanism was obtained with a single modeling framework that unifies the modeling results obtained in existing studies, and the product quality and amount were predicted with good accuracy. The results on prediction error propagation show that, although the models are data-driven, the design of the plansifter model can be done independently of that of the roller mill model, since the prediction error did not propagate across the models strongly.

Mismatch negativity indexes illness-specific impairments of cortical plasticity in schizophrenia: A comparison with bipolar disorder and Alzheimer's disease

Cognitive impairment is an important predictor of functional outcome in patients with schizophrenia, yet its neurobiology is still incompletely understood. Neuropathological evidence of impaired synaptic connectivity and NMDA receptor-dependent transmission in superior temporal cortex motivated us to explore the correlation of in vivo mismatch negativity (MMN) with cognitive status in patients with schizophrenia. MMN elicited in a roving stimulus paradigm displayed a response proportional to the number of stimulus repetitions (memory trace effect). Preliminary evidence in patients with chronic schizophrenia suggests that attenuation of this MMN memory trace effect was correlated with the degree of neuropsychological memory dysfunction. Here we present data from a larger confirmatory study in patients with schizophrenia, bipolar disorder, probable Alzheimer's disease and healthy controls. We observed that the diminution of the MMN memory trace effect and its correlation with memory impairment was only found in the schizophrenia group. Recent pharmacological studies using the roving paradigm suggest that attenuation of the MMN trace effect can be understood as abnormal modulation of NMDA receptor-dependent plasticity. We suggest that the convergence of the previously identified synaptic pathology in supragranular cortical layers with the intracortical locus of MMN generation accounts for the remarkable robustness of MMN impairments in schizophrenia. We further speculate that this layer-specific synaptic pathology identified in supragranular neurons plays a pivotal computational role, by weakening the encoding and propagation of prediction errors to higher cortical modules. According to predictive coding theory such breakdown will have grave implications not only for perception, but also for higher-order cognition and may thus account for the MMN–cognition correlations observed here. Finally, MMN is a sensitive and specific biomarker for detecting the early prodromal phase of schizophrenia and is well suited for the exploration of novel cognition-enhancing agents in humans.

Set membership prediction of nonlinear time series

In this paper, a prediction method for nonlinear time series based on a set membership (SM) approach is proposed. The method does not require the choice of the functional form of the model used for prediction, but assumes a bound on the rate of variation of the regression function defining the model. At the contrary, most of the existing prediction methods need the choice of a functional form of the regression function or of state equations (piecewise linear, quadratic, etc.) and this choice is usually the result of heuristic searches. These searches may be quite time consuming, and lead only to approximate model structures, whose errors may be responsible of bad propagation of prediction errors, especially for the multistep ahead prediction. Moreover, the method proposed in this paper assumes only that the noise is bounded, in contrast with statistical approaches, which rely on noise assumptions such as stationarity, ergodicity, uncorrelation, type of distribution, etc. The validity of these assumptions may be difficult to be reliably tested in many applications and is certainly lost in presence of approximate modeling. In the present SM approach, using a result developed in a previous paper, the values of the bounds on the gradient of the regression function and on the noise can be suitably assessed to verify the validity tests. Two almost optimal prediction algorithms are then derived, the second one having improved optimal properties over the first one, at the expense of an increased computational complexity. The method is tested and compared with other literature methods on the well-known Wolf Sunspot Numbers series, widely used in the time series literature as a benchmark test, and on the prediction of vertical dynamics of vehicles with controlled suspensions. A simulation example is also presented to investigate how much conservative the SM approach may be in the most adverse situation where data are generated by a linear autoregressive (AR) model driven by i.i.d. gaussian white noise and the SM prediction is compared with the optimal statistical predictor, which makes use of the exact assumptions.

Uncertainty, identifiability and the propagation of prediction errors: A case study of lake ontario

AbstractMost models of environmental systems can be expected to suffer, to a varying extent, from a lack of identifiability. Overparameterization seems both intrinsic and an intractable problem. As a result, interpretation of past observed behaviour through a model is ambiguous, and this will have obvious implications for any predictions of future behaviour. There has been much study of this issue of uncertainty in environmental modelling, yet very limited investigation of its consequences in the use of larger‐scale models. The paper examines some of these consequences for a model of the carbon and phosphorus cycles in Lake Ontario having 24 state variables and thirteen parameters. Filtering theory is used to provide both the conceptual framework for the analysis and a specific, recursive algorithm for the propagation of a variance‐covariance matrix of state prediction errors. Special attention is given to examining the ambiguities that may arise in the making of predictions, and the possibility of generating contradictory statements about future behaviour is introduced. It is found that with this relatively large model the propagation of prediction error variances is insensitive to the degree of correlation among the parameter‐estimation errors and to changes of the nominal estimated values of small subsets of parameters.

Water quality modeling: A review of the analysis of uncertainty

This paper reviews the role of uncertainty in the identification of mathematical models of water quality and in the application of these models to problems of prediction. More specifically, four problem areas are examined in detail: uncertainty about model structure, uncertainty in the estimated model parameter values, the propagation of prediction errors, and the design of experiments in order to reduce the critical uncertainties associated with a model. The review is rather lengthy, and it has therefore been prepared in effect as two papers. There is a shorter, largely nontechnical version, which gives a quick impression of the current and future issues in the analysis of uncertainty in water quality modeling. Enclosed by this shorter discussion is the main body of the review dealing in turn with (1) identifiability and experimental design, (2) the generation of preliminary model hypotheses under conditions of sparse, grossly uncertain field data, (3) the selection and evaluation of model structure, (4) parameter estimation (model calibration), (5) checks and balances on the identified model, i.e., model “verification” and model discrimination, and (6) prediction error propagation. Much time is spent in discussing the algorithms of system identification, in particular, the methods of recursive estimation, and in relating these algorithms and the subject of identification to the problems of prediction uncertainty and first‐order error analysis. There are two obvious omissions from the review. It is not concerned primarily with either the development and solution of stochastic differential equations or the issue of decision making under uncertainty, although clearly some reference must be made to these topics. In brief, the review concludes (not surprisingly) that much work has been done on the analysis of uncertainty in the development of mathematical models of water quality, and much remains to be done. A lack of model identifiability has been an outstanding difficulty in the interpretation and explanation of past observed system behavior, and there is ample evidence to show that the “larger,” more “comprehensive” models are easily capable of generating highly uncertain predictions of future behavior. For the future of the subject, it is speculated that there is the possibility of progress in the development of novel algorithms for model structure identification, a need for new questions to be posed in the problem of prediction, and a distinct challenge to the conventional views of this review in the new forms of knowledge representation and manipulation now emerging from the field of artificial intelligence.