End-to-end neural system identification with neural information flow

Neural information flow (NIF) provides a novel approach for system identification in neuroscience. It models the neural computations in multiple brain regions and can be trained end-to-end via stochastic gradient descent from noninvasive data. NIF models represent neural information processing via a network of coupled tensors, each encoding the representation of the sensory input contained in a brain region. The elements of these tensors can be interpreted as cortical columns whose activity encodes the presence of a specific feature in a spatiotemporal location. Each tensor is coupled to the measured data specific to a brain region via low-rank observation models that can be decomposed into the spatial, temporal and feature receptive fields of a localized neuronal population. Both these observation models and the convolutional weights defining the information processing within regions are learned end-to-end by predicting the neural signal during sensory stimulation. We trained a NIF model on the activity of early visual areas using a large-scale fMRI dataset recorded in a single participant. We show that we can recover plausible visual representations and population receptive fields that are consistent with empirical findings.     

Pruning artificial neural networks using neural complexity measures

This paper describes a new method for pruning artificial neural networks, using a measure of the neural complexity of the neural network. This measure is used to determine the connections that should be pruned. The measure computes the information-theoretic complexity of a neural network, which is similar to, yet different from previous research on pruning. The method proposed here shows how overly large and complex networks can be reduced in size, whilst retaining learnt behaviour and fitness. The technique proposed here helps to discover a network topology that matches the complexity of the problem it is meant to solve. This novel pruning technique is tested in a robot control domain, simulating a racecar. It is shown, that the proposed pruning method is a significant improvement over the most commonly used pruning method Magnitude Based Pruning. Furthermore, some of the pruned networks prove to be faster learners than the benchmark network that they originate from. This means that this pruning method can also help to unleash hidden potential in a network, because the learning time decreases substantially for a pruned a network, due to the reduction of dimensionality of the network.     

Markovian neural networks

The neural network that efficiently and nearly optimally solves difficult optimization problems is defined. The convergence proof for the Markovian neural network that asynchronously updates its neurons' states is also presented. The comparison of the performance of the Markovian neural network with various combinatorial optimization methods in two domains is described. The Markovian neural network is shown to be an efficient tool for solving optimization problems.     

Noise-induced neural impulses

The firing pattern of neural pulses often show the following features: the shapes of individual pulses are nearly identical and frequency independent; the firing frequency can vary over a broad range; the time period between pulses shows a stochastic scatter. This behaviour cannot be understood on the basis of a deterministic non-linear dynamic process, e.g. the Bonhoeffer-van der Pol model. We demonstrate in this paper that a noise term added to the Bonhoeffer-van der Pol model can reproduce the firing patterns of neurons very well. For this purpose we have considered the Fokker-Planck equation corresponding to the stochastic Bonhoeffer-van der Pol model. This equation has been solved by a new Monte Carlo algorithm. We demonstrate that the ensuing distribution functions represent only the global characteristics of the underlying force field: lines of zero slope which attract nearby trajectories prove to be the regions of phase space where the distributions concentrate their amplitude. Since there are two such lines the distributions are bimodal representing repeated fluctuations between two lines of zero slope. Even in cases where the deterministic Bonhoeffer-van der Pol model does not show limit cycle behaviour the stochastic system produces a limit cycle. This cycle can be identified with the firing of neural pulses.     

Cephalopod neural networks

Cephalopods have arguably the largest and most complex nervous systems amongst the invertebrates; but despite the squid giant axon being one of the best studied nerve cells in neuroscience, and the availability of superb information on the morphology of some cephalopod brains, there is surprisingly little known about the operation of the neural networks that underlie the sophisticated range of behaviour these animals display. This review focuses on a few of the best studied neural networks: the giant fiber system, the chromatophore system, the statocyst system, the visual system and the learning and memory system, with a view to summarizing our current knowledge and stimulating new studies, particularly on the activities of identified central neurons, to provide a more complete understanding of networks within the cephalopod nervous system.     

Global rhythmic activities in hippocampal neural fields and neural coding

Global oscillations of the neural field represent some of the most interesting expressions of the hippocampal activity, being related also to learning and memory. To study oscillatory activities of the CA3 field in theta range, a model of this sub-field of Hippocampus has been formulated. The model describes the firing activity of CA3 neuronal populations within the frame of a kinetic theory of neural systems and it has been used for computer simulations. The results show that the propagation of activities induced in the neural field by hippocampal afferents occurs only in narrow time windows confined by inhibitory barrages, whose time-course follows the theta rhythm. Moreover, during each period of a theta wave, the entire CA3 field bears a firing activity with peculiar space-time patterns, a sort of specific imprint, which can induce effects with similar patterns on brain regions driven by the hippocampal formation. The simulation has also demonstrated the ability of medial septum to influence the global activity of the CA3 pyramidal population through the control of the population of inhibitory interneurons. At last, the possible involvement of global population oscillations in neural coding has been discussed.     

Mouse-chick neural chimeras

Embryonic chimera production was used to study the developmental processes of the mouse nervous system. The difficulty of performing in situ transplantation experiments of neural primordium of mouse embryo was overcome by isotopic and isochronic grafting of mouse neural tube fragments into chick embryo. Mouse neural tube cells differentiated perfectly in ovo and neural crest cells associated with the grafted neural tube were able to migrate and reach the normal arrest sites of host neural crests. Cranial neural crest cells penetrated into chick facial areas and entered into the development of dental bud structures, participating in vibrissa formation. Depending on graft level, in ovo implanted mouse neural crest cells formed different components of the peripheral nervous system. At trunk level, they located in spinal ganglia and orthosympathetic chains and gave rise to Schwann cells lining the nerves. When implanted into the lumbosacral region, they penetrated into the enteric nervous system. At the precise 18-24 somite level, they colonized host adrenal gland. Mouse neural tube was involved in the mechanisms required to maintain myogenesis in host somites. Furthermore in ovo grafts of mouse cells from genetically modified embryos, in which many mutations induce early death, are particularly useful to investigate cellular events involved in the development of the nervous system and to identify molecular events of embryogenesis.     

Drosophila neural pathways

Summary TheDrosophila giant axon pathways cervical connective — thoracic indirect flight muscles were studied by a combined electrophysiological and genetic analysis. A functional coupling of the left and right giant axon pathways was revealed by intracellular recordings of electrical responses of the thoracic indirect flight muscles, when evoked by electrical stimulation of cervical connective (Fig. 2). This functional coupling was demonstrated in wild-type flies and in flies of the single gene, temperature-sensitive paralytic mutation,para ^ts . The functional coupling was evident also in selected bilateral gynandromorph flies, mosaics for thepara ^ts mutation (Fig. 1), even at restricted elevated ambient temperature (Tables 1–3). Analysis of neurally evoked electrogenic muscle responses of wild-type flies, following injection of picrotoxin, verifies the notion that both the dorsoventral and the dorsolongitudinal flight muscles share a common activating pathway (Fig. 3). Picrotoxin application to gynandromorph flies demonstrated the existence of neuronal elements additional to the giant axon pathways, that evoke the indirect flight muscles in response to cervical stimulation (Figs. 4, 5). An unexpected finding was the poor correlation between the mosaic external phenotype of the gynandromorph flies ofpara ^ts mutation and the genotype of neural pathways activating their thoracic flight muscles, as evidenced by the intracellular recordings.Abbreviations GA giant axon - DVM dorsoventral muscle - DLM dorsolongitudinal muscle - PSI peripherally synapsing interneuron - ts temperature sensitive     

Bayesian neural networks

A neural network that uses the basic Hebbian learning rule and the Bayesian combination function is defined. Analogously to Hopfield's neural network, the convergence for the Bayesian neural network that asynchronously updates its neurons' states is proved. The performance of the Bayesian neural network in four medical domains is compared with various classification methods. The Bayesian neural network uses more sophisticated combination function than Hopfield's neural network and uses more economically the available information. The naive Bayesian classifier typically outperforms the basic Bayesian neural network since iterations in network make too many mistakes. By restricting the number of iterations and increasing the number of fixed points the network performs better than the naive Bayesian classifier. The Bayesian neural network is designed to learn very quickly and incrementally.     

Cardiac neural crest

Neural crest cells (NCCs) contribute to many organs and tissues during embryonic development. Amongst these, the cardiovascular system represents a fascinating example. In this review, recent advances in our understanding of the developmental biology and molecular genetics regulating cardiac NCC maturation will be summarized. While the existence of a significant neural crest (NC) contribution to the developing heart has been appreciated for more than 20 years, only in the last few years have molecular pathways regulating this process been elucidated and the significant contribution of these mechanisms to the etiology of congenital heart disease in man become apparent. Emerging data suggest that ongoing studies will reveal complex inductive interactions between cardiac NC and a series of other cell types contributing to the developing cardiovascular system.     

Microfabricated nerve–electrode interfaces in neural prosthetics and neural engineering

Neural interfaces and implants are finding more clinical applications and there are rapid technological advances for more efficient and safe design, fabrication and materials to establish high-fidelity neural interfaces. In this review paper, we highlight new developments of the microfabricated electrodes and substrates with regard to the design, materials, fabrication and their clinical applications. There is a noticeable trend towards integration of microfluidic modules on a single neural platform. In addition to the microelectrodes for neural recording and stimulation, microfluidic channels are integrated into a nerve–electrode interface to explore the rich neurochemistry present at the neural interface and exploit it for enhanced electrochemical stimulation and recording of the central and peripheral nervous system.     

Interpreting neural computations by examining intrinsic and embedding dimensionality of neural activity

The ongoing exponential rise in recording capacity calls for new approaches for analysing and interpreting neural data. Effective dimensionality has emerged as an important property of neural activity across populations of neurons, yet different studies rely on different definitions and interpretations of this quantity. Here, we focus on intrinsic and embedding dimensionality, and discuss how they might reveal computational principles from data. Reviewing recent works, we propose that the intrinsic dimensionality reflects information about the latent variables encoded in collective activity while embedding dimensionality reveals the manner in which this information is processed. We conclude by highlighting the role of network models as an ideal substrate for testing more specifically various hypotheses on the computational principles reflected through intrinsic and embedding dimensionality.