Stimulus statistics shape oscillations in nonlinear recurrent neural networks.

Rhythmic activity plays a central role in neural computations and brain functions ranging from homeostasis to attention, as well as in neurological and neuropsychiatric disorders. Despite this pervasiveness, little is known about the mechanisms whereby the frequency and power of oscillatory activity are modulated, and how they reflect the inputs received by neurons. Numerous studies have reported input-dependent fluctuations in peak frequency and power (as well as couplings across these features). However, it remains unresolved what mediates these spectral shifts among neural populations. Extending previous findings regarding stochastic nonlinear systems and experimental observations, we provide analytical insights regarding oscillatory responses of neural populations to stimulation from either endogenous or exogenous origins. Using a deceptively simple yet sparse and randomly connected network of neurons, we show how spiking inputs can reliably modulate the peak frequency and power expressed by synchronous neural populations without any changes in circuitry. Our results reveal that a generic, non-nonlinear and input-induced mechanism can robustly mediate these spectral fluctuations, and thus provide a framework in which inputs to the neurons bidirectionally regulate both the frequency and power expressed by synchronous populations. Theoretical and computational analysis of the ensuing spectral fluctuations was found to reflect the underlying dynamics of the input stimuli driving the neurons. Our results provide insights regarding a generic mechanism supporting spectral transitions observed across cortical networks and spanning multiple frequency bands.

Correlations between negative-induced cerebral dynamics and idiopathic intellectual disability: an EEG study

Intellectual disability has long been associated with deficits in socio-emotional processing. However, studies investigating brain dynamics of maladaptive socio-emotional skills associated with intellectual disability are scarce. Here, we compared differences in brain activity between low intelligence quotient (I.Q.<75, N=13) and normal controls (N=15) while evaluating their subjective emotions. Positive (P) and negative (N) valenced pictures were presented one at a time to participants of both groups, at a rate of ¾. The task required that each participant evaluate their subjective emotion and press a predefined push-button when done, alternatively P and N. Electroencephalographic (EEG) signals were continuously recorded, and the 1000ms time window following each picture was analyzed offline for power in frequency domain. Alpha low (8-10Hz) and upper (10-13Hz) frequency bands were then compared for both groups and for both P and N emotions in 12 distributed scalp electrodes. The qualitative evaluation of emotions was similar between both groups, with constant longer reaction times for the low IQ participants. The EEG signal comparison shows marked power decrease in upper alpha frequency range for N emotions in low intelligence group. Otherwise no significant difference was noticed between low and normal IQ. Main findings of the present study are (1) results do not support the hypothesis that impairment in developmental intelligence roots in maladaptive emotional processing; (2) the strong alpha power suppression during negative-induced emotions suggests the involvement of an extended neural network and more effortful inhibition processes than positive ones. We call for further studies with a larger sample.

Neural plasticity associated with recently versus often heard objects.

In natural settings the same sound source is often heard repeatedly, with variations in spectro-temporal and spatial characteristics. We investigated how such repetitions influence sound representations and in particular how auditory cortices keep track of recently vs. often heard objects. A set of 40 environmental sounds was presented twice, i.e. as prime and as repeat, while subjects categorized the corresponding sound sources as living vs. non-living. Electrical neuroimaging analyses were applied to auditory evoked potentials (AEPs) comparing primes vs. repeats (effect of presentation) and the four experimental sections. Dynamic analysis of distributed source estimations revealed i) a significant main effect of presentation within the left temporal convexity at 164-215ms post-stimulus onset; and ii) a significant main effect of section in the right temporo-parietal junction at 166-213ms. A 3-way repeated measures ANOVA (hemisphere×presentation×section) applied to neural activity of the above clusters during the common time window confirmed the specificity of the left hemisphere for the effect of presentation, but not that of the right hemisphere for the effect of section. In conclusion, spatio-temporal dynamics of neural activity encode the temporal history of exposure to sound objects. Rapidly occurring plastic changes within the semantic representations of the left hemisphere keep track of objects heard a few seconds before, independent of the more general sound exposure history. Progressively occurring and more long-lasting plastic changes occurring predominantly within right hemispheric networks, which are known to code for perceptual, semantic and spatial aspects of sound objects, keep track of multiple exposures.

Brain dynamics of meal size selection in humans.

Although neuroimaging research has evidenced specific responses to visual food stimuli based on their nutritional quality (e.g., energy density, fat content), brain processes underlying portion size selection remain largely unexplored. We identified spatio-temporal brain dynamics in response to meal images varying in portion size during a task of ideal portion selection for prospective lunch intake and expected satiety. Brain responses to meal portions judged by the participants as 'too small', 'ideal' and 'too big' were measured by means of electro-encephalographic (EEG) recordings in 21 normal-weight women. During an early stage of meal viewing (105-145ms), data showed an incremental increase of the head-surface global electric field strength (quantified via global field power; GFP) as portion judgments ranged from 'too small' to 'too big'. Estimations of neural source activity revealed that brain regions underlying this effect were located in the insula, middle frontal gyrus and middle temporal gyrus, and are similar to those reported in previous studies investigating responses to changes in food nutritional content. In contrast, during a later stage (230-270ms), GFP was maximal for the 'ideal' relative to the 'non-ideal' portion sizes. Greater neural source activity to 'ideal' vs. 'non-ideal' portion sizes was observed in the inferior parietal lobule, superior temporal gyrus and mid-posterior cingulate gyrus. Collectively, our results provide evidence that several brain regions involved in attention and adaptive behavior track 'ideal' meal portion sizes as early as 230ms during visual encounter. That is, responses do not show an increase paralleling the amount of food viewed (and, in extension, the amount of reward), but are shaped by regulatory mechanisms.

Cooperative and Competitive Spreading Dynamics on the Human Connectome.

Increasingly detailed data on the network topology of neural circuits create a need for theoretical principles that explain how these networks shape neural communication. Here we use a model of cascade spreading to reveal architectural features of human brain networks that facilitate spreading. Using an anatomical brain network derived from high-resolution diffusion spectrum imaging (DSI), we investigate scenarios where perturbations initiated at seed nodes result in global cascades that interact either cooperatively or competitively. We find that hub regions and a backbone of pathways facilitate early spreading, while the shortest path structure of the connectome enables cooperative effects, accelerating the spread of cascades. Finally, competing cascades become integrated by converging on polysensory associative areas. These findings show that the organizational principles of brain networks shape global communication and facilitate integrative function.

Nature of the Epidemic Threshold for the Susceptible-Infected-Susceptible Dynamics in Networks

We develop an analytical approach to the susceptible-infected-susceptible epidemic model that allows us to unravel the true origin of the absence of an epidemic threshold in heterogeneous networks. We find that a delicate balance between the number of high degree nodes in the network and the topological distance between them dictates the existence or absence of such a threshold. In particular, small-world random networks with a degree distribution decaying slower than an exponential have a vanishing epidemic threshold in the thermodynamic limit.

The Armc10/SVH gene: Genome context, regulation of mitochondrial dynamics and protection against Aβ-induced mitochondrial fragmentation

Mitochondrial function and dynamics are essential for neurotransmission, neural function and neuronal viability. Recently, we showed that the eutherian-specific Armcx gene cluster (Armcx1-6 genes), located in the X chromosome, encodes for a new family of proteins that localise to mitochondria, regulating mitochondrial trafficking. The Armcx gene cluster evolved by retrotransposition of the Armc10 gene mRNA, which is present in all vertebrates and is considered to be the ancestor gene. Here we investigate the genomic organisation, mitochondrial functions and putative neuroprotective role of the Armc10 ancestor gene. The genomic context of the Armc10 locus shows considerable syntenic conservation among vertebrates, and sequence comparisons and CHIP-data suggest the presence of at least three conserved enhancers. We also show that the Armc10 protein localises to mitochondria and that it is highly expressed in the brain. Furthermore, we show that Armc10 levels regulate mitochondrial trafficking in neurons, but not mitochondrial aggregation, by controlling the number of moving mitochondria. We further demonstrate that the Armc10 protein interacts with the KIF5/Miro1-2/Trak2 trafficking complex. Finally, we show that overexpression of Armc10 in neurons prevents A beta-induced mitochondrial fission and neuronal death. Our data suggest both conserved and differential roles of the Armc10/Armcx gene family in regulating mitochondrial dynamics in neurons, and underscore a protective effect of the Armc10 gene against A beta-induced toxicity. Overall, our findings support a further degree of regulation of mitochondrial dynamics in the brain of more evolved mammals.

The Armc10/SVH gene: Genome context, regulation of mitochondrial dynamics and protection against Aβ-induced mitochondrial fragmentation

Mitochondrial function and dynamics are essential for neurotransmission, neural function and neuronal viability. Recently, we showed that the eutherian-specific Armcx gene cluster (Armcx1-6 genes), located in the X chromosome, encodes for a new family of proteins that localise to mitochondria, regulating mitochondrial trafficking. The Armcx gene cluster evolved by retrotransposition of the Armc10 gene mRNA, which is present in all vertebrates and is considered to be the ancestor gene. Here we investigate the genomic organisation, mitochondrial functions and putative neuroprotective role of the Armc10 ancestor gene. The genomic context of the Armc10 locus shows considerable syntenic conservation among vertebrates, and sequence comparisons and CHIP-data suggest the presence of at least three conserved enhancers. We also show that the Armc10 protein localises to mitochondria and that it is highly expressed in the brain. Furthermore, we show that Armc10 levels regulate mitochondrial trafficking in neurons, but not mitochondrial aggregation, by controlling the number of moving mitochondria. We further demonstrate that the Armc10 protein interacts with the KIF5/Miro1-2/Trak2 trafficking complex. Finally, we show that overexpression of Armc10 in neurons prevents A beta-induced mitochondrial fission and neuronal death. Our data suggest both conserved and differential roles of the Armc10/Armcx gene family in regulating mitochondrial dynamics in neurons, and underscore a protective effect of the Armc10 gene against A beta-induced toxicity. Overall, our findings support a further degree of regulation of mitochondrial dynamics in the brain of more evolved mammals.

Brain dynamics sustaining rapid rule extraction from speech

Language acquisition is a complex process that requires the synergic involvement of different cognitive functions, which include extracting and storing the words of the language and their embedded rules for progressive acquisition of grammatical information. As has been shown in other fields that study learning processes, synchronization mechanisms between neuronal assemblies might have a key role during language learning. In particular, studying these dynamics may help uncover whether different oscillatory patterns sustain more item-based learning of words and rule-based learning from speech input. Therefore, we tracked the modulation of oscillatory neural activity during the initial exposure to an artificial language, which contained embedded rules. We analyzed both spectral power variations, as a measure of local neuronal ensemble synchronization, as well as phase coherence patterns, as an index of the long-range coordination of these local groups of neurons. Synchronized activity in the gamma band (2040 Hz), previously reported to be related to the engagement of selective attention, showed a clear dissociation of local power and phase coherence between distant regions. In this frequency range, local synchrony characterized the subjects who were focused on word identification and was accompanied by increased coherence in the theta band (48 Hz). Only those subjects who were able to learn the embedded rules showed increased gamma band phase coherence between frontal, temporal, and parietal regions.

Flexible multibody dynamics and intelligent control of a hydraulically driven hybrid redundant robot machine

The assembly and maintenance of the International Thermonuclear Experimental Reactor (ITER) vacuum vessel (VV) is highly challenging since the tasks performed by the robot involve welding, material handling, and machine cutting from inside the VV. The VV is made of stainless steel, which has poor machinability and tends to work harden very rapidly, and all the machining operations need to be carried out from inside of the ITER VV. A general industrial robot cannot be used due to its poor stiffness in the heavy duty machining process, and this will cause many problems, such as poor surface quality, tool damage, low accuracy. Therefore, one of the most suitable options should be a light weight mobile robot which is able to move around inside of the VV and perform different machining tasks by replacing different cutting tools. Reducing the mass of the robot manipulators offers many advantages: reduced material costs, reduced power consumption, the possibility of using smaller actuators, and a higher payload-to-robot weight ratio. Offsetting these advantages, the lighter weight robot is more flexible, which makes it more difficult to control. To achieve good machining surface quality, the tracking of the end effector must be accurate, and an accurate model for a more flexible robot must be constructed. This thesis studies the dynamics and control of a 10 degree-of-freedom (DOF) redundant hybrid robot (4-DOF serial mechanism and 6-DOF 6-UPS hexapod parallel mechanisms) hydraulically driven with flexible rods under the influence of machining forces. Firstly, the flexibility of the bodies is described using the floating frame of reference method (FFRF). A finite element model (FEM) provided the Craig-Bampton (CB) modes needed for the FFRF. A dynamic model of the system of six closed loop mechanisms was assembled using the constrained Lagrange equations and the Lagrange multiplier method. Subsequently, the reaction forces between the parallel and serial parts were used to study the dynamics of the serial robot. A PID control based on position predictions was implemented independently to control the hydraulic cylinders of the robot. Secondly, in machining, to achieve greater end effector trajectory tracking accuracy for surface quality, a robust control of the actuators for the flexible link has to be deduced. This thesis investigates the intelligent control of a hydraulically driven parallel robot part based on the dynamic model and two schemes of intelligent control for a hydraulically driven parallel mechanism based on the dynamic model: (1) a fuzzy-PID self-tuning controller composed of the conventional PID control and with fuzzy logic, and (2) adaptive neuro-fuzzy inference system-PID (ANFIS-PID) self-tuning of the gains of the PID controller, which are implemented independently to control each hydraulic cylinder of the parallel mechanism based on rod length predictions. The serial component of the hybrid robot can be analyzed using the equilibrium of reaction forces at the universal joint connections of the hexa-element. To achieve precise positional control of the end effector for maximum precision machining, the hydraulic cylinder should be controlled to hold the hexa-element. Thirdly, a finite element approach of multibody systems using the Special Euclidean group SE(3) framework is presented for a parallel mechanism with flexible piston rods under the influence of machining forces. The flexibility of the bodies is described using the nonlinear interpolation method with an exponential map. The equations of motion take the form of a differential algebraic equation on a Lie group, which is solved using a Lie group time integration scheme. The method relies on the local description of motions, so that it provides a singularity-free formulation, and no parameterization of the nodal variables needs to be introduced. The flexible slider constraint is formulated using a Lie group and used for modeling a flexible rod sliding inside a cylinder. The dynamic model of the system of six closed loop mechanisms was assembled using Hamilton’s principle and the Lagrange multiplier method. A linearized hydraulic control system based on rod length predictions was implemented independently to control the hydraulic cylinders. Consequently, the results of the simulations demonstrating the behavior of the robot machine are presented for each case study. In conclusion, this thesis studies the dynamic analysis of a special hybrid (serialparallel) robot for the above-mentioned special task involving the ITER and investigates different control algorithms that can significantly improve machining performance. These analyses and results provide valuable insight into the design and control of the parallel robot with flexible rods.

Modeling High-Dimensional Audio Sequences with Recurrent Neural Networks

Cette thèse étudie des modèles de séquences de haute dimension basés sur des réseaux de neurones récurrents (RNN) et leur application à la musique et à la parole. Bien qu'en principe les RNN puissent représenter les dépendances à long terme et la dynamique temporelle complexe propres aux séquences d'intérêt comme la vidéo, l'audio et la langue naturelle, ceux-ci n'ont pas été utilisés à leur plein potentiel depuis leur introduction par Rumelhart et al. (1986a) en raison de la difficulté de les entraîner efficacement par descente de gradient. Récemment, l'application fructueuse de l'optimisation Hessian-free et d'autres techniques d'entraînement avancées ont entraîné la recrudescence de leur utilisation dans plusieurs systèmes de l'état de l'art. Le travail de cette thèse prend part à ce développement. L'idée centrale consiste à exploiter la flexibilité des RNN pour apprendre une description probabiliste de séquences de symboles, c'est-à-dire une information de haut niveau associée aux signaux observés, qui en retour pourra servir d'à priori pour améliorer la précision de la recherche d'information. Par exemple, en modélisant l'évolution de groupes de notes dans la musique polyphonique, d'accords dans une progression harmonique, de phonèmes dans un énoncé oral ou encore de sources individuelles dans un mélange audio, nous pouvons améliorer significativement les méthodes de transcription polyphonique, de reconnaissance d'accords, de reconnaissance de la parole et de séparation de sources audio respectivement. L'application pratique de nos modèles à ces tâches est détaillée dans les quatre derniers articles présentés dans cette thèse. Dans le premier article, nous remplaçons la couche de sortie d'un RNN par des machines de Boltzmann restreintes conditionnelles pour décrire des distributions de sortie multimodales beaucoup plus riches. Dans le deuxième article, nous évaluons et proposons des méthodes avancées pour entraîner les RNN. Dans les quatre derniers articles, nous examinons différentes façons de combiner nos modèles symboliques à des réseaux profonds et à la factorisation matricielle non-négative, notamment par des produits d'experts, des architectures entrée/sortie et des cadres génératifs généralisant les modèles de Markov cachés. Nous proposons et analysons également des méthodes d'inférence efficaces pour ces modèles, telles la recherche vorace chronologique, la recherche en faisceau à haute dimension, la recherche en faisceau élagué et la descente de gradient. Finalement, nous abordons les questions de l'étiquette biaisée, du maître imposant, du lissage temporel, de la régularisation et du pré-entraînement.

Forecasting Global Temperature Variations by Neural Networks

Global temperature variations between 1861 and 1984 are forecast usingsregularization networks, multilayer perceptrons and linearsautoregression. The regularization network, optimized by stochasticsgradient descent associated with colored noise, gives the bestsforecasts. For all the models, prediction errors noticeably increasesafter 1965. These results are consistent with the hypothesis that thesclimate dynamics is characterized by low-dimensional chaos and thatsthe it may have changed at some point after 1965, which is alsosconsistent with the recent idea of climate change.s

Wnt6 controls amniote neural crest induction through the non-canonical signaling pathway

The neural crest is a multipotent embryonic cell population that arises from neural ectoderm and forms derivatives essential for vertebrate function. Neural crest induction requires an ectodermal signal, thought to be a Writ ligand, but the identity of the Wnt that performs this function in amniotes is unknown. Here, we demonstrate that Wnt6, derived from the ectoderm, is necessary for chick neural crest induction. Crucially, we also show that Wnt6 acts through the non-canonical pathway and not the beta-catenin-dependant pathway. Surprisingly, we found that canonical Wnt signaling inhibited neural crest production in the chick embryo. In light of studies in anamniotes demonstrating that canonical Wnt signaling induces neural crest, these results indicate a significant and novel change in the mechanism of neural crest induction during vertebrate evolution. These data also highlight a key role for noncanonical Wnt signaling in cell type specification from a stem population during development.

Neural network enhanced generalised minimum variance self-tuning controller for nonlinear discrete-time systems

A neural network enhanced self-tuning controller is presented, which combines the attributes of neural network mapping with a generalised minimum variance self-tuning control (STC) strategy. In this way the controller can deal with nonlinear plants, which exhibit features such as uncertainties, nonminimum phase behaviour, coupling effects and may have unmodelled dynamics, and whose nonlinearities are assumed to be globally bounded. The unknown nonlinear plants to be controlled are approximated by an equivalent model composed of a simple linear submodel plus a nonlinear submodel. A generalised recursive least squares algorithm is used to identify the linear submodel and a layered neural network is used to detect the unknown nonlinear submodel in which the weights are updated based on the error between the plant output and the output from the linear submodel. The procedure for controller design is based on the equivalent model therefore the nonlinear submodel is naturally accommodated within the control law. Two simulation studies are provided to demonstrate the effectiveness of the control algorithm.