Investigation of neural correlates of faces and emotional expressions using magnetoencephalography

The recognition of faces and of facial expressions in an important evolutionary skill, and an integral part of social communication. It has been argued that the processing of faces is distinct from the processing of non-face stimuli and functional neuroimaging investigations have even found evidence of a distinction between the perception of faces and of emotional expressions. Structural and temporal correlates of face perception and facial affect have only been separately identified. Investigation neural dynamics of face perception per se as well as facial affect would allow the mapping of these in space, time and frequency specific domains. Participants were asked to perform face categorisation and emotional discrimination tasks and Magnetoencephalography (MEG) was used to measure the neurophysiology of face and facial emotion processing. SAM analysis techniques enable the investigation of spectral changes within specific time-windows and frequency bands, thus allowing the identification of stimulus specific regions of cortical power changes. Furthermore, MEG’s excellent temporal resolution allows for the detection of subtle changes associated with the processing of face and non-face stimuli and different emotional expressions. The data presented reveal that face perception is associated with spectral power changes within a distributed cortical network comprising occipito-temporal as well as parietal and frontal areas. For the perception of facial affect, spectral power changes were also observed within frontal and limbic areas including the parahippocampal gyrus and the amygdala. Analyses of temporal correlates also reveal a distinction between the processing of faces and facial affect. Face perception per se occurred at earlier latencies whereas the discrimination of facial expression occurred within a longer time-window. In addition, the processing of faces and facial affect was differentially associated with changes in cortical oscillatory power for alpha, beta and gamma frequencies. The perception of faces and facial affect is associated with distinct changes in cortical oscillatory activity that can be mapped to specific neural structures, specific time-windows and latencies as well as specific frequency bands. Therefore, the work presented in this thesis provides further insight into the sequential processing of faces and facial affect.

Description and training of neural network dynamics

Attractor properties of a popular discrete-time neural network model are illustrated through numerical simulations. The most complex dynamics is found to occur within particular ranges of parameters controlling the symmetry and magnitude of the weight matrix. A small network model is observed to produce fixed points, limit cycles, mode-locking, the Ruelle-Takens route to chaos, and the period-doubling route to chaos. Training algorithms for tuning this dynamical behaviour are discussed. Training can be an easy or difficult task, depending whether the problem requires the use of temporal information distributed over long time intervals. Such problems require training algorithms which can handle hidden nodes. The most prominent of these algorithms, back propagation through time, solves the temporal credit assignment problem in a way which can work only if the relevant information is distributed locally in time. The Moving Targets algorithm works for the more general case, but is computationally intensive, and prone to local minima.

Dynamics of on-line gradient descent learning for multilayer neural networks

We consider the problem of on-line gradient descent learning for general two-layer neural networks. An analytic solution is presented and used to investigate the role of the learning rate in controlling the evolution and convergence of the learning process.

The learning dynamics of a universal approximator

The learning properties of a universal approximator, a normalized committee machine with adjustable biases, are studied for on-line back-propagation learning. Within a statistical mechanics framework, numerical studies show that this model has features which do not exist in previously studied two-layer network models without adjustable biases, e.g., attractive suboptimal symmetric phases even for realizable cases and noiseless data.

Finite-size effects in on-line learning of multilayer neural networks

We complement recent advances in thermodynamic limit analyses of mean on-line gradient descent learning dynamics in multi-layer networks by calculating fluctuations possessed by finite dimensional systems. Fluctuations from the mean dynamics are largest at the onset of specialisation as student hidden unit weight vectors begin to imitate specific teacher vectors, increasing with the degree of symmetry of the initial conditions. In light of this, we include a term to stimulate asymmetry in the learning process, which typically also leads to a significant decrease in training time.

Dynamics of on-line learning in radial basis function networks

On-line learning is examined for the radial basis function network, an important and practical type of neural network. The evolution of generalization error is calculated within a framework which allows the phenomena of the learning process, such as the specialization of the hidden units, to be analyzed. The distinct stages of training are elucidated, and the role of the learning rate described. The three most important stages of training, the symmetric phase, the symmetry-breaking phase, and the convergence phase, are analyzed in detail; the convergence phase analysis allows derivation of maximal and optimal learning rates. As well as finding the evolution of the mean system parameters, the variances of these parameters are derived and shown to be typically small. Finally, the analytic results are strongly confirmed by simulations.

The dynamics of matrix momentum

We analyse the matrix momentum algorithm, which provides an efficient approximation to on-line Newton's method, by extending a recent statistical mechanics framework to include second order algorithms. We study the efficacy of this method when the Hessian is available and also consider a practical implementation which uses a single example estimate of the Hessian. The method is shown to provide excellent asymptotic performance, although the single example implementation is sensitive to the choice of training parameters. We conjecture that matrix momentum could provide efficient matrix inversion for other second order algorithms.

Dynamics of learning with restricted training sets

The dynamics of supervised learning in layered neural networks were studied in the regime where the size of the training set is proportional to the number of inputs. The evolution of macroscopic observables, including the two relevant performance measures can be predicted by using the dynamical replica theory. Three approximation schemes aimed at eliminating the need to solve a functional saddle-point equation at each time step have been derived.

Imaging the dynamics of the auditory steady-state evoked response

This study used magnetoencephalography (MEG) to examine the dynamic patterns of neural activity underlying the auditory steady-state response. We examined the continuous time-series of responses to a 32-Hz amplitude modulation. Fluctuations in the amplitude of the evoked response were found to be mediated by non-linear interactions with oscillatory processes both at the same source, in the alpha and beta frequency bands, and in the opposite hemisphere. © 2005 Elsevier Ireland Ltd. All rights reserved.

Dynamics and topographic organization of recursive self-organizing maps

Recently there has been an outburst of interest in extending topographic maps of vectorial data to more general data structures, such as sequences or trees. However, there is no general consensus as to how best to process sequences using topographicmaps, and this topic remains an active focus of neurocomputational research. The representational capabilities and internal representations of the models are not well understood. Here, we rigorously analyze a generalization of the self-organizingmap (SOM) for processing sequential data, recursive SOM (RecSOM) (Voegtlin, 2002), as a nonautonomous dynamical system consisting of a set of fixed input maps. We argue that contractive fixed-input maps are likely to produce Markovian organizations of receptive fields on the RecSOM map. We derive bounds on parameter β (weighting the importance of importing past information when processing sequences) under which contractiveness of the fixed-input maps is guaranteed. Some generalizations of SOM contain a dynamic module responsible for processing temporal contexts as an integral part of the model. We show that Markovian topographic maps of sequential data can be produced using a simple fixed (nonadaptable) dynamic module externally feeding a standard topographic model designed to process static vectorial data of fixed dimensionality (e.g., SOM). However, by allowing trainable feedback connections, one can obtain Markovian maps with superior memory depth and topography preservation. We elaborate on the importance of non-Markovian organizations in topographic maps of sequential data. © 2006 Massachusetts Institute of Technology.

The spatial distribution and temporal dynamics of brain regions activated during the perception of object and non-object patterns

Both animal and human studies suggest that the efficiency with which we are able to grasp objects is attributable to a repertoire of motor signals derived directly from vision. This is in general agreement with the long-held belief that the automatic generation of motor signals by the perception of objects is based on the actions they afford. In this study, we used magnetoencephalography (MEG) to determine the spatial distribution and temporal dynamics of brain regions activated during passive viewing of object and non-object targets that varied in the extent to which they afforded a grasping action. Synthetic Aperture Magnetometry (SAM) was used to localize task-related oscillatory power changes within specific frequency bands, and the time course of activity within given regions-of-interest was determined by calculating time-frequency plots using a Morlet wavelet transform. Both single subject and group-averaged data on the spatial distribution of brain activity are presented. We show that: (i) significant reductions in 10-25 Hz activity within extrastriate cortex, occipito-temporal cortex, sensori-motor cortex and cerebellum were evident with passive viewing of both objects and non-objects; and (ii) reductions in oscillatory activity within the posterior part of the superior parietal cortex (area Ba7) were only evident with the perception of objects. Assuming that focal reductions in low-frequency oscillations (< 30 Hz) reflect areas of heightened neural activity, we conclude that: (i) activity within a network of brain areas, including the sensori-motor cortex, is not critically dependent on stimulus type and may reflect general changes in visual attention; and (ii) the posterior part of the superior parietal cortex, area Ba7, is activated preferentially by objects and may play a role in computations related to grasping. © 2006 Elsevier Inc. All rights reserved.

A novel silicon membrane-based biosensing platform using distributive sensing strategy and artificial neural networks for feature analysis

A novel biosensing system based on a micromachined rectangular silicon membrane is proposed and investigated in this paper. A distributive sensing scheme is designed to monitor the dynamics of the sensing structure. An artificial neural network is used to process the measured data and to identify cell presence and density. Without specifying any particular bio-application, the investigation is mainly concentrated on the performance testing of this kind of biosensor as a general biosensing platform. The biosensing experiments on the microfabricated membranes involve seeding different cell densities onto the sensing surface of membrane, and measuring the corresponding dynamics information of each tested silicon membrane in the form of a series of frequency response functions (FRFs). All of those experiments are carried out in cell culture medium to simulate a practical working environment. The EA.hy 926 endothelial cell lines are chosen in this paper for the bio-experiments. The EA.hy 926 endothelial cell lines represent a particular class of biological particles that have irregular shapes, non-uniform density and uncertain growth behaviour, which are difficult to monitor using the traditional biosensors. The final predicted results reveal that the methodology of a neural-network based algorithm to perform the feature identification of cells from distributive sensory measurement has great potential in biosensing applications.

Evolutionary neural networks : models and applications

The scaling problems which afflict attempts to optimise neural networks (NNs) with genetic algorithms (GAs) are disclosed. A novel GA-NN hybrid is introduced, based on the bumptree, a little-used connectionist model. As well as being computationally efficient, the bumptree is shown to be more amenable to genetic coding lthan other NN models. A hierarchical genetic coding scheme is developed for the bumptree and shown to have low redundancy, as well as being complete and closed with respect to the search space. When applied to optimising bumptree architectures for classification problems the GA discovers bumptrees which significantly out-perform those constructed using a standard algorithm. The fields of artificial life, control and robotics are identified as likely application areas for the evolutionary optimisation of NNs. An artificial life case-study is presented and discussed. Experiments are reported which show that the GA-bumptree is able to learn simulated pole balancing and car parking tasks using only limited environmental feedback. A simple modification of the fitness function allows the GA-bumptree to learn mappings which are multi-modal, such as robot arm inverse kinematics. The dynamics of the 'geographic speciation' selection model used by the GA-bumptree are investigated empirically and the convergence profile is introduced as an analytical tool. The relationships between the rate of genetic convergence and the phenomena of speciation, genetic drift and punctuated equilibrium arc discussed. The importance of genetic linkage to GA design is discussed and two new recombination operators arc introduced. The first, linkage mapped crossover (LMX) is shown to be a generalisation of existing crossover operators. LMX provides a new framework for incorporating prior knowledge into GAs.Its adaptive form, ALMX, is shown to be able to infer linkage relationships automatically during genetic search.

Modulation of long-range neural synchrony reflects temporal limitations of visual attention in humans

Because of attentional limitations, the human visual system can process for awareness and response only a fraction of the input received. Lesion and functional imaging studies have identified frontal, temporal, and parietal areas as playing a major role in the attentional control of visual processing, but very little is known about how these areas interact to form a dynamic attentional network. We hypothesized that the network communicates by means of neural phase synchronization, and we used magnetoencephalography to study transient long-range interarea phase coupling in a well studied attentionally taxing dual-target task (attentional blink). Our results reveal that communication within the fronto-parieto-temporal attentional network proceeds via transient long-range phase synchronization in the beta band. Changes in synchronization reflect changes in the attentional demands of the task and are directly related to behavioral performance. Thus, we show how attentional limitations arise from the way in which the subsystems of the attentional network interact. The human brain faces an inestimable task of reducing a potentially overloading amount of input into a manageable flow of information that reflects both the current needs of the organism and the external demands placed on it. This task is accomplished via a ubiquitous construct known as “attention,” whose mechanism, although well characterized behaviorally, is far from understood at the neurophysiological level. Whereas attempts to identify particular neural structures involved in the operation of attention have met with considerable success (1-5) and have resulted in the identification of frontal, parietal, and temporal regions, far less is known about the interaction among these structures in a way that can account for the task-dependent successes and failures of attention. The goal of the present research was, thus, to unravel the means by which the subsystems making up the human attentional network communicate and to relate the temporal dynamics of their communication to observed attentional limitations in humans. A prime candidate for communication among distributed systems in the human brain is neural synchronization (for review, see ref. 6). Indeed, a number of studies provide converging evidence that long-range interarea communication is related to synchronized oscillatory activity (refs. 7-14; for review, see ref. 15). To determine whether neural synchronization plays a role in attentional control, we placed humans in an attentionally demanding task and used magnetoencephalography (MEG) to track interarea communication by means of neural synchronization. In particular, we presented 10 healthy subjects with two visual target letters embedded in streams of 13 distractor letters, appearing at a rate of seven per second. The targets were separated in time by a single distractor. This condition leads to the “attentional blink” (AB), a well studied dual-task phenomenon showing the reduced ability to report the second of two targets when an interval <500 ms separates them (16-18). Importantly, the AB does not prevent perceptual processing of missed target stimuli but only their conscious report (19), demonstrating the attentional nature of this effect and making it a good candidate for the purpose of our investigation. Although numerous studies have investigated factors, e.g., stimulus and timing parameters, that manipulate the magnitude of a particular AB outcome, few have sought to characterize the neural state under which “standard” AB parameters produce an inability to report the second target on some trials but not others. We hypothesized that the different attentional states leading to different behavioral outcomes (second target reported correctly or not) are characterized by specific patterns of transient long-range synchronization between brain areas involved in target processing. Showing the hypothesized correspondence between states of neural synchronization and human behavior in an attentional task entails two demonstrations. First, it needs to be demonstrated that cortical areas that are suspected to be involved in visual-attention tasks, and the AB in particular, interact by means of neural synchronization. This demonstration is particularly important because previous brain-imaging studies (e.g., ref. 5) only showed that the respective areas are active within a rather large time window in the same task and not that they are concurrently active and actually create an interactive network. Second, it needs to be demonstrated that the pattern of neural synchronization is sensitive to the behavioral outcome; specifically, the ability to correctly identify the second of two rapidly succeeding visual targets

Anisotropic opinion dynamics

We consider the process of opinion formation in a society of interacting agents, where there is a set B of socially accepted rules. In this scenario, we observed that agents, represented by simple feed-forward, adaptive neural networks, may have a conservative attitude (mostly in agreement with B) or liberal attitude (mostly in agreement with neighboring agents) depending on how much their opinions are influenced by their peers. The topology of the network representing the interaction of the society's members is determined by a graph, where the agents' properties are defined over the vertexes and the interagent interactions are defined over the bonds. The adaptability of the agents allows us to model the formation of opinions as an online learning process, where agents learn continuously as new information becomes available to the whole society (online learning). Through the application of statistical mechanics techniques we deduced a set of differential equations describing the dynamics of the system. We observed that by slowly varying the average peer influence in such a way that the agents attitude changes from conservative to liberal and back, the average social opinion develops a hysteresis cycle. Such hysteretic behavior disappears when the variance of the social influence distribution is large enough. In all the cases studied, the change from conservative to liberal behavior is characterized by the emergence of conservative clusters, i.e., a closed knitted set of society members that follow a leader who agrees with the social status quo when the rule B is challenged.