Higher-order Representation Research Articles

High-Level Prediction Signals in a Low-Level Area of the Macaque Face-Processing Hierarchy

Theories like predictive coding propose that lower-order brain areas compare their inputs to predictions derived from higher-order representations and signal their deviation as a prediction error. Here, we investigate whether the macaque face-processing system, a three-level hierarchy in the ventral stream, employs such a coding strategy. We show that after statistical learning of specific face sequences, the lower-level face area ML computes the deviation of actual from predicted stimuli. But these signals do not reflect the tuning characteristic of ML. Rather, they exhibit identity specificity and view invariance, the tuning properties of higher-level face areas AL and AM. Thus, learning appears to endow lower-level areas with the capability to test predictions at a higher level of abstraction than what is afforded by the feedforward sweep. These results provide evidence for computational architectures like predictive coding and suggest a new quality of functional organization of information-processing hierarchies beyond pure feedforward schemes.

Monge–Ampére simulation of fourth order PDEs in two dimensions with application to elastic–electrostatic contact problems

We present an efficient moving mesh method for the simulation of fourth order nonlinear partial differential equations (PDEs) in two dimensions using the Parabolic Monge–Ampére (PMA) equation. PMA methods have been successfully applied to the simulation of second order problems, but not on systems with higher order equations which arise in many topical applications. Our main application is the resolution of fine scale behavior in PDEs describing elastic–electrostatic interactions. The PDE system considered has multiple parameter dependent singular solution modalities, including finite time singularities and sharp interface dynamics. We describe how to construct a dynamic mesh algorithm for such problems which incorporates known self similar or boundary layer scalings of the underlying equation to locate and dynamically resolve fine scale solution features in these singular regimes. We find a key step in using the PMA equation for mesh generation in fourth order problems is the adoption of a high order representation of the transformation from the computational to physical mesh. We demonstrate the efficacy of the new method on a variety of examples and establish several new results and conjectures on the nature of self-similar singularity formation in higher order PDEs.

Evidence for cue-independent spatial representation in the human auditory cortex during active listening

Few auditory functions are as important or as universal as the capacity for auditory spatial awareness (e.g., sound localization). That ability relies on sensitivity to acoustical cues-particularly interaural time and level differences (ITD and ILD)-that correlate with sound-source locations. Under nonspatial listening conditions, cortical sensitivity to ITD and ILD takes the form of broad contralaterally dominated response functions. It is unknown, however, whether that sensitivity reflects representations of the specific physical cues or a higher-order representation of auditory space (i.e., integrated cue processing), nor is it known whether responses to spatial cues are modulated by active spatial listening. To investigate, sensitivity to parametrically varied ITD or ILD cues was measured using fMRI during spatial and nonspatial listening tasks. Task type varied across blocks where targets were presented in one of three dimensions: auditory location, pitch, or visual brightness. Task effects were localized primarily to lateral posterior superior temporal gyrus (pSTG) and modulated binaural-cue response functions differently in the two hemispheres. Active spatial listening (location tasks) enhanced both contralateral and ipsilateral responses in the right hemisphere but maintained or enhanced contralateral dominance in the left hemisphere. Two observations suggest integrated processing of ITD and ILD. First, overlapping regions in medial pSTG exhibited significant sensitivity to both cues. Second, successful classification of multivoxel patterns was observed for both cue types and-critically-for cross-cue classification. Together, these results suggest a higher-order representation of auditory space in the human auditory cortex that at least partly integrates the specific underlying cues.

Exposure to complex environments results in more sparse representations of space in the hippocampus.

The neural circuitry mediating sensory and motor representations is adaptively tuned by an animal's interaction with its environment. Similarly, higher order representations such as spatial memories can be modified by exposure to a complex environment (CE), but in this case the changes in brain circuitry that mediate the effect are less well understood. Here, we show that prolonged CE exposure was associated with increased selectivity of CA1 "place cells" to a particular recording arena compared to a social control (SC) group. Furthermore, fewer CA1 and DG neurons in the CE group expressed high levels of Arc protein, a marker of recent activation, following brief exposure to a completely novel environment. The reduced Arc expression was not attributable to overall changes in cell density or number. These data indicate that one effect of CE exposure is to modify high-level spatial representations in the brain by increasing the sparsity of population coding within networks of neurons. Greater sparsity could result in a more efficient and compact coding system that might alter behavioural performance on spatial tasks. The results from a behavioural experiment were consistent with this hypothesis, as CE-treated animals habituated more rapidly to a novel environment despite showing equivalent initial responding.

Striatal activity is modulated by target probability.

Target probability has well-known neural effects. In the brain, target probability is known to affect frontal activity, with lower probability targets producing more prefrontal activation than those that occur with higher probability. Although the effect of target probability on cortical activity is well specified, its effect on subcortical structures such as the striatum is less well understood. Here, I examined this issue and found that the striatum was highly responsive to target probability. This is consistent with its hypothesized role in the gating of salient information into higher-order task representations. The current data are interpreted in light of that fact that different components of the striatum are sensitive to different types of task-relevant information.

Action observation activates neurons of the monkey ventrolateral prefrontal cortex

Prefrontal cortex is crucial for exploiting contextual information for the planning and guidance of behavioral responses. Among contextual cues, those provided by others’ behavior are particularly important, in primates, for selecting appropriate reactions and suppressing the inappropriate ones. These latter functions deeply rely on the ability to understand others’ actions. However, it is largely unknown whether prefrontal neurons are activated by action observation. To address this issue, we recorded the activity of ventrolateral prefrontal (VLPF) neurons of macaque monkeys during the observation of videos depicting biological movements performed by a monkey or a human agent, and object motion. Our results show that a population of VLPF neurons respond to the observation of biological movements, in particular those representing goal directed actions. Many of these neurons also show a preference for the agent performing the action. The neural response is present also when part of the observed movement is obscured, suggesting that these VLPF neurons code a high order representation of the observed action rather than a simple visual description of it.

Optimal numerical solvers for transient simulations of ice flow using the Ice Sheet System Model (ISSM versions 4.2.5 and 4.11)

Abstract. Identifying fast and robust numerical solvers is a critical issue that needs to be addressed in order to improve projections of polar ice sheets evolving in a changing climate. This work evaluates the impact of using advanced numerical solvers for transient ice-flow simulations conducted with the JPL–UCI Ice Sheet System Model (ISSM). We identify optimal numerical solvers by testing a broad suite of readily available solvers, ranging from direct sparse solvers to preconditioned iterative methods, on the commonly used Ice Sheet Model Intercomparison Project for Higher-Order ice sheet Models benchmark tests. Three types of analyses are considered: mass transport, horizontal stress balance, and incompressibility. The results of the fastest solvers for each analysis type are ranked based on their scalability across mesh size and basal boundary conditions. We find that the fastest iterative solvers are ∼ 1.5–100 times faster than the default direct solver used in ISSM, with speed-ups improving rapidly with increased mesh resolution. We provide a set of recommendations for users in search of efficient solvers to use for transient ice-flow simulations, enabling higher-resolution meshes and faster turnaround time. The end result will be improved transient simulations for short-term, highly resolved forward projections (10–100 year time scale) and also improved long-term paleo-reconstructions using higher-order representations of stresses in the ice. This analysis will also enable a new generation of comprehensive uncertainty quantification assessments of forward sea-level rise projections, which rely heavily on ensemble or sampling approaches that are inherently expensive.

A high-order representation and classification method for transcription factor binding sites recognition in Escherichia coli.

Identifying transcription factors binding sites (TFBSs) plays an important role in understanding gene regulatory processes. The underlying mechanism of the specific binding for transcription factors (TFs) is still poorly understood. Previous machine learning-based approaches to identifying TFBSs commonly map a known TFBS to a one-dimensional vector using its physicochemical properties. However, when the dimension-sample rate is large (i.e., number of dimensions/number of samples), concatenating different physicochemical properties to a one-dimensional vector not only is likely to lose some structural information, but also poses significant challenges to recognition methods. In this paper, we introduce a purely geometric representation method, tensor (also called multidimensional array), to represent TFs using their physicochemical properties. Accompanying the multidimensional array representation, we also develop a tensor-based recognition method, tensor partial least squares classifier (abbreviated as TPLSC). Intuitively, multidimensional arrays enable borrowing more information than one-dimensional arrays. The performance of each method is evaluated by average F-measure on 51 Escherichia coli TFs from RegulonDB database. In our first experiment, the results show that multiple nucleotide properties can obtain more power than dinucleotide properties. In the second experiment, the results demonstrate that our method can gain increased prediction power, roughly 33% improvements more than the best result from existing methods. The representation method for TFs is an important step in TFBSs recognition. We illustrate the benefits of this representation on real data application via a series of experiments. This method can gain further insights into the mechanism of TF binding and be of great use for metabolic engineering applications.

Concurrent use of somatotopic and external reference frames in a tactile mislocalization task

Localizing tactile stimuli on our body requires sensory information to be represented in multiple frames of reference along the sensory pathways. These reference frames include the representation of sensory information in skin coordinates, in which the spatial relationship of skin regions is maintained. The organization of the primary somatosensory cortex matches such somatotopic reference frame. In contrast, higher-order representations are based on external coordinates, in which body posture and gaze direction are taken into account in order to localise touch in other meaningful ways according to task demands. Dominance of one representation or the other, or the use of multiple representations with different weights, is thought to depend on contextual factors of cognitive and/or sensory origins. However, it is unclear under which situations a reference frame takes over another or when different reference frames are jointly used at the same time. The study of tactile mislocalizations at the fingers has shown a key role of the somatotopic frame of reference, both when touches are delivered unilaterally to a single hand, and when they are delivered bilaterally to both hands. Here, we took advantage of a well-established tactile mislocalization paradigm to investigate whether the reference frame used to integrate bilateral tactile stimuli can change as a function of the spatial relationship between the two hands. Specifically, supra-threshold interference stimuli were applied to the index or little fingers of the left hand 200ms prior to the application of a test stimulus on a finger of the right hand. Crucially, different hands postures were adopted (uncrossed or crossed). Results show that introducing a change in hand-posture triggered the concurrent use of somatotopic and external reference frames when processing bilateral touch at the fingers. This demonstrates that both somatotopic and external reference frames can be concurrently used to localise tactile stimuli on the fingers.

Uncertain Systems Order Reduction by Modal Analysis Approach

Objectives: Modeling of physical systems results in complex higher-order representation. It’s somewhat strenuous to process out with these intricate systems, in such conditions these large scale systems are approximated by relegated order model. Methods/Statistical Analysis: A system with bounded parameters but uncertain and having constant coefficients is termed as an interval system. In this jotting Modal analysis approach, order reduction technique has been used to demote the higher order system to its relegated order. Numerical examples are solved to show the supremacy of this advanced technique. Findings: By using the proposed method, the step response of original and reduced order uncertain systems are closer when compared to other methods. The relative integral square error values are also less as compared to other techniques. Application/Improvements: The relegated model acquired by this approach has its behavior homogeneous to the original system. The stability is vowed if the original system is stable. In order to depiction a controller for a higher order system it is quite arduous, so by using the order demotion technique it becomes more facile.Keywords: Kharitonov’s Theorem, Modal Analysis Approach, Order Reduction and Stability Uncertain Systems.

Method for Higher Order polynomial Sugeno Fuzzy Inference Systems

A method for Higher Order polynomial Sugeno Fuzzy Inference Systems formation is presented. Compared to other existing Higher Order Sugeno implementations, it uses fewer parameters; and compared to Zero and 1st Order Sugeno Fuzzy Systems it has overall improved model performance. While best models are not always obtained via a Higher Order representation, in our proposed method it is possible to choose the polynomial Order which best fits the desired data. Its input is a previously established model found by a clustering algorithm (subtractive algorithm in this case). Afterward, parameters of all Higher Order polynomials are adjusted using Recursive Least Square algorithm. For experimental validation, multiple benchmark datasets are tested using Hold-Out and K-fold validation as well as data forecasting. Various performance measures are used, although Akaike Information Criterion is used as a primary measure to demonstrate that our proposed Higher Order polynomials have overall better model performance over Zero and 1st Order polynomials.

Consciousness and Theory of Mind: a Common Theory?

Many philosophers and scientists have argued that the difference between phenomenally conscious states and other kind of states lies in the implicit self-awareness that conscious states have. Higher-Order Representationalist (HOR) theories, attempt to explain such a self-awareness by means of a higher-order representation. Consciousness relies on our capacity to represent our own mental states, consciousness depends on our Theory of Mind. Such an ability can, at least conceptually, be decomposed into another two: mindreading and metacognition. In this paper I will argue that consciousness cannot depend on mindreading. The tenability of HOR theories depends, therefore, on the relation between mindreading and metacognition. I analyze several views on such a relation and argue that none of them seem to be a plausible option for HOR theories.

A local and global statistics pattern analysis method and its application to process fault identification

Traditional principal component analysis (PCA) is a second-order method and lacks the ability to provide higher-order representations for data variables. Recently, a statistics pattern analysis (SPA) framework has been incorporated into PCA model to make full use of various statistics of data variables effectively. However, these methods omit the local information, which is also important for process monitoring and fault diagnosis. In this paper, a local and global statistics pattern analysis (LGSPA) method, which integrates SPA framework and locality preserving projections within the PCA, is proposed to utilize various statistics and preserve both local and global information in the observed data. For the purpose of fault detection, two monitoring indices are constructed based on the LGSPA model. In order to identify fault variables, an improved reconstruction based contribution (IRBC) plot based on LGSPA model is proposed to locate fault variables. The RBC of various statistics of original process variables to the monitoring indices is calculated with the proposed RBC method. Based on the calculated RBC of process variables' statistics, a new contribution of process variables is built to locate fault variables. The simulation results on a simple six-variable system and a continuous stirred tank reactor system demonstrate that the proposed fault diagnosis method can effectively detect fault and distinguish the fault variables from normal variables.

Tools to detect structures in dynamical systems using Jet Transport

This paper is devoted to the development of some dynamical indicators that allow the determination of regions and structures that separate different dynamic regimes in autonomous and non-autonomous dynamical systems. The underlying idea is closely related to the Lagrangian coherent structures concept introduced by Haller. In the present paper, instead of using the Cauchy–Green tensor, that determines the domains where the flow associated to a differential equation is expanding in the normal direction, the Jet Transport methodology is used. This is a semi-numerical tool, that has as basic ingredients a polynomial algebra package and a numerical integration method, allowing, at each integration step, the propagation under a flow of a neighbourhood U instead of a single initial condition. The output of the procedure is a polynomial in several variables that represents the image of U up to a selected order, containing high order terms of the variational equations. Using these high order representation, the places where the normal direction expands can be easily detected, in a similar manner as the procedures for calculating the Lagrangian coherent structures do. In order to illustrate the methodology, first the results obtained in the determination of the separatrices of the simple and the periodically perturbed pendulum are given. Later, the applications to the circular restricted three body problem are considered, where the aim is the detection of invariant manifolds of libration point orbits, as well as in the non-autonomous vector field defined by the elliptic restricted three body problem.

Visual search in depth: The neural correlates of segmenting a display into relevant and irrelevant three-dimensional regions

Visual perception is facilitated by the ability to selectively attend to relevant parts of the world and to ignore irrelevant regions or features. In visual search tasks, viewers are able to segment displays into relevant and irrelevant items based on a number of factors including the colour, motion, and temporal onset of the target and distractors. Understanding the process by which viewers prioritise relevant parts of a display can provide insights into the effect of top-down control on visual perception. Here, we investigate the behavioural and neural correlates of segmenting a display according to the expected three-dimensional (3D) location of a target. We ask whether this segmentation is based on low-level visual features (e.g. common depth or common surface) or on higher-order representations of 3D regions. Similar response-time benefits and neural activity were obtained when items fell on common surfaces or within depth-defined volumes, and when displays were vertical (such that items shared a common depth/disparity) or were tilted in depth. These similarities indicate that segmenting items according to their 3D location is based on attending to a 3D region, rather than a specific depth or surface. Segmenting the items in depth was mainly associated with increased activation in depth-sensitive parietal regions rather than in depth-sensitive visual regions. We conclude that segmenting items in depth is primarily achieved via higher-order, cue invariant representations rather than through filtering in lower-level perceptual regions.

Sparse Deep Stacking Network for Image Classification

Sparse coding can learn good robust representation to noise and model more higher-order representation for image classification. However, the inference algorithm is computationally expensive even though the supervised signals are used to learn compact and discriminative dictionaries in sparse coding techniques. Luckily, a simplified neural network module (SNNM) has been proposed to directly learn the discriminative dictionaries for avoiding the expensive inference. But the SNNM module ignores the sparse representations. Therefore, we propose a sparse SNNM module by adding the mixed-norm regularization (l1/l2 norm). The sparse SNNM modules are further stacked to build a sparse deep stacking network (S-DSN). In the experiments, we evaluate S-DSN with four databases, including Extended YaleB, AR, 15 scene and Caltech101. Experimental results show that our model outperforms related classification methods with only a linear classifier. It is worth noting that we reach 98.8% recognition accuracy on 15 scene.

Behavioral relevance helps untangle natural vocal categories in a specific subset of core auditory cortical pyramidal neurons.

Sound categorization is essential for auditory behaviors like acoustic communication, but its genesis within the auditory pathway is not well understood-especially for learned natural categories like vocalizations, which often share overlapping acoustic features that must be distinguished (e.g., speech). We use electrophysiological mapping and single-unit recordings in mice to investigate how representations of natural vocal categories within core auditory cortex are modulated when one category acquires enhanced behavioral relevance. Taking advantage of a maternal mouse model of acoustic communication, we found no long-term auditory cortical map expansion to represent a behaviorally relevant pup vocalization category-contrary to expectations from the cortical plasticity literature on conditioning with pure tones. Instead, we observed plasticity that improved the separation between acoustically similar pup and adult vocalization categories among a physiologically defined subset of late-onset, putative pyramidal neurons, but not among putative interneurons. Additionally, a larger proportion of these putative pyramidal neurons in maternal animals compared with nonmaternal animals responded to the individual pup call exemplars having combinations of acoustic features most typical of that category. Together, these data suggest that higher-order representations of acoustic categories arise from a subset of core auditory cortical pyramidal neurons that become biased toward the combination of acoustic features statistically predictive of membership to a behaviorally relevant sound category.

A learning rule that explains how rewards teach attention

Many theories propose that top-down attentional signals control processing in sensory cortices by modulating neural activity. But who controls the controller? Here we investigate how a biologically plausible neural reinforcement learning scheme can create higher order representations and top-down attentional signals. The learning scheme trains neural networks using two factors that gate Hebbian plasticity: (1) an attentional feedback signal from the response-selection stage to earlier processing levels; and (2) a globally available neuromodulator that encodes the reward prediction error. We demonstrate how the neural network learns to direct attention to one of two coloured stimuli that are arranged in a rank-order. Like monkeys trained on this task, the network develops units that are tuned to the rank-order of the colours and it generalizes this newly learned rule to previously unseen colour combinations. These results provide new insight into how individuals can learn to control attention as a function of reward contingency.

THE TIME-HARMONIC DISCONTINUOUS GALERKIN METHOD AS A ROBUST FORWARD SOLVER FOR MICROWAVE IMAGING APPLICATIONS

Novel microwave imaging systems require flexible forward solvers capable of incorporating arbitrary boundary conditions and inhomogeneous background constitutive parameters. In this work we focus on the implementation of a time-harmonic Discontinuous Galerkin Method (DGM) forward solver with a number of features that aim to benefit tomographic microwave imaging algorithms: locally varying high-order polynomial field expansions, locally varying high-order representations of the complex constitutive parameters, and exact radiating boundary conditions. The DGM formulated directly from Maxwell's curl equations facilitates including both electric and magnetic contrast functions, the latter being important when considering quantitative imaging with magnetic contrast agents. To improve forward solver performance we formulate the DGM for time-harmonic electric and magnetic vector wave equations driven by both electric and magnetic sources. Sufficient implementation details are provided to permit existing DGM codes based on nodal expansions of Maxwell's curl equations to be converted to the wave equation formulations. Results are shown to validate the DGM forward solver framework for transverse magnetic problems that might typically be found in tomographic imaging systems, illustrating how high-order expansions of the constitutive parameters can be used to improve forward solver performance.

Differentiating intended sensory outcome from underlying motor actions in the human brain.

To achieve a certain sensory outcome, multiple actions can be executed. For example, unlocking a door might require clockwise or counterclockwise key turns depending on regional norms. Using fMRI in healthy human subjects, we examined the neural networks that dissociate intended sensory outcome from underlying motor actions. Subjects controlled a figure on a computer screen by performing pen traces on an MR-compatible digital tablet. Our design allowed us to dissociate intended sensory outcome (moving the figure in a certain direction) from the underlying motor action (horizontal/vertical pen traces). Using multivoxel pattern analysis and a whole-brain searchlight strategy, we found that activity patterns in left (contralateral) motor and parietal cortex and also right (ipsilateral) motor cortex significantly discriminated direction of pen traces regardless of intended direction of figure movement. Conversely, activity patterns in right superior parietal lobule and premotor cortex, and also left frontopolar cortex, significantly discriminated intended direction of figure movement regardless of underlying direction of hand movement. Together, these results highlight the role of ipsilateral motor cortex in coding movement directions and point to a network of brain regions involved in high order representation of intended sensory outcome that is dissociated from specific motor plans.