A long-term goal of tissue engineering is to exploit the ability of supporting materials to govern cell-specific behaviors. Instructive scaffolds code such information by modulating (via their physical and chemical features) the interface between cells and materials at the nanoscale. In modern neuroscience, therapeutic regenerative strategies (i.e., brain repair after damage) aim to guide and enhance the intrinsic capacity of the brain to reorganize by promoting plasticity mechanisms in a controlled fashion. Direct and specific interactions between synthetic materials and biological cell membranes may play a central role in this process. Here, we investigate the role of the material's properties alone, in carbon nanotube scaffolds, in constructing the functional building blocks of neural circuits: the synapses. Using electrophysiological recordings and rat cultured neural networks, we describe the ability of a nanoscaled material to promote the formation of synaptic contacts and to modulate their plasticity.
Normalization is a critical operation in neural circuits. In the brain, there is evidence that normalization is implemented via inhibitory interneurons and allows neural populations to adjust to changes in the distribution of their inputs. In artificial neural networks (ANNs), normalization is used to improve learning in tasks that involve complex input distributions. However, it is unclear whether inhibition-mediated normalization in biological neural circuits also improves learning. Here, we explore this possibility using ANNs with separate excitatory and inhibitory populations trained on an image recognition task with variable luminosity. We find that inhibition-mediated normalization does not improve learning if normalization is applied only during inference. However, when this normalization is extended to include back-propagated errors, performance improves significantly. These results suggest that if inhibition-mediated normalization improves learning in the brain, it additionally requires the normalization of learning signals.
There have been two recent revolutionary advances in neuroscience: First, genetically encoded activity sensors have brought the goal of optical detection of single action potentials in vivo within reach. Second, optogenetic actuators now allow the activity of neurons to be controlled with millisecond precision. These revolutions have now been combined, together with advanced microscopies, to allow "all-optical" readout and manipulation of activity in neural circuits with single-spike and single-neuron precision. This is a transformational advance that will open new frontiers in neuroscience research. Harnessing the power of light in the all-optical approach requires coexpression of genetically encoded activity sensors and optogenetic probes in the same neurons, as well as the ability to simultaneously target and record the light from the selected neurons. It has recently become possible to combine sensors and optical strategies that are sufficiently sensitive and cross talk free to enable single-action-potential sensitivity and precision for both readout and manipulation in the intact brain. The combination of simultaneous readout and manipulation from the same genetically defined cells will enable a wide range of new experiments as well as inspire new technologies for interacting with the brain. The advances described in this review herald a future where the traditional tools used for generations by physiologists to study and interact with the brain-stimulation and recording electrodes-can largely be replaced by light. We outline potential future developments in this field and discuss how the all-optical strategy can be applied to solve fundamental problems in neuroscience. SIGNIFICANCE STATEMENT: This review describes the nexus of dramatic recent developments in optogenetic probes, genetically encoded activity sensors, and novel microscopies, which together allow the activity of neural circuits to be recorded and manipulated entirely using light. The optical and protein engineering strategies that form the basis of this "all-optical" approach are now sufficiently advanced to enable single-neuron and single-action potential precision for simultaneous readout and manipulation from the same functionally defined neurons in the intact brain. These advances promise to illuminate many fundamental challenges in neuroscience, including transforming our search for the neural code and the links between neural circuit activity and behavior.
Converging evidence from diverse studies suggests that atypical brain connectivity in autism affects in distinct ways short- and long-range cortical pathways, disrupting neural communication and the balance of excitation and inhibition. This hypothesis is based mostly on functional non-invasive studies that show atypical synchronization and connectivity patterns between cortical areas in children and adults with autism. Indirect methods to study the course and integrity of major brain pathways at low resolution show changes in fractional anisotropy (FA) or diffusivity of the white matter in autism. Findings in post-mortem brains of adults with autism provide evidence of changes in the fine structure of axons below prefrontal cortices, which communicate over short- or long-range pathways with other cortices and subcortical structures. Here we focus on evidence of cellular and axon features that likely underlie the changes in short- and long-range communication in autism. We review recent findings of changes in the shape, thickness, and volume of brain areas, cytoarchitecture, neuronal morphology, cellular elements, and structural and neurochemical features of individual axons in the white matter, where pathology is evident even in gross images. We relate cellular and molecular features to imaging and genetic studies that highlight a variety of polymorphisms and epigenetic factors that primarily affect neurite growth and synapse formation and function in autism. We report preliminary findings of changes in autism in the ratio of distinct types of inhibitory neurons in prefrontal cortex, known to shape network dynamics and the balance of excitation and inhibition. Finally we present a model that synthesizes diverse findings by relating them to developmental events, with a goal to identify common processes that perturb development in autism and affect neural communication, reflected in altered patterns of attention, social interactions, and language.
Mixed-signal neuromorphic systems represent a promising solution for solving extreme-edge computing tasks without relying on external computing resources. Their spiking neural network circuits are optimized for processing sensory data on-line in continuous-time. However, their low precision and high variability can severely limit their performance. To address this issue and improve their robustness to inhomogeneities and noise in both their internal state variables and external input signals, we designed on-chip learning circuits with short-term analog dynamics and long-term tristate discretization mechanisms. An additional hysteretic stop-learning mechanism is included to improve stability and automatically disable weight updates when necessary, to enable continuous always-on learning. We designed a spiking neural network with these learning circuits in a prototype chip using a 180 nm CMOS technology. Simulation and silicon measurement results from the prototype chip are presented. These circuits enable the construction of large-scale spiking neural networks with online learning capabilities for real-world edge computing tasks.
The interpretation of the origin of observed exoplanets is usually done only qualitatively due to uncertainties of key parameters in planet formation models. To allow a quantitative methodology which traces back in time to the planet birth locations, we train recently developed conditional invertible neural networks (cINN) on synthetic data from a global planet formation model which tracks growth from dust grains to evolved final giant planets. In addition to deterministic single planet formation runs, we also include gravitationally interacting planets in multiplanetary systems, which include some measure of chaos. For the latter case, we treat them as individual planets or choose the two or three planets most likely to be discovered by telescopes. We find that training on multiplanetary data, each planet treated as individual point, is promising. The single-planet data only covers a small range of planets and does not extrapolate well to planet properties not included in the training data. Extension to planetary systems will require more training data due to the higher dimensionality of the problem.
This paper presents an ML-driven framework for automated RF physical synthesis that transforms circuit netlists into manufacturable GDSII layouts. While recent ML approaches demonstrate success in topology selection and parameter optimization, they fail to produce manufacturable layouts due to oversimplified component models and lack of routing capabilities. Our framework addresses these limitations through three key innovations: (1) a neural network framework trained on 18,210 inductor geometries with frequency sweeps from 1-100 GHz, generating 7.5 million training samples, that predicts inductor Q-factor with less than 2% error and enables fast gradient-based layout optimization with a 93.77% success rate in producing high-Q layouts; (2) an intelligent P-Cell optimizer that reduces layout area while maintaining design-rule-check (DRC) compliance; and (3) a complete placement and routing engine with frequency-dependent EM spacing rules and DRC-aware synthesis. The neural inductor model demonstrates superior accuracy across 1-100 GHz, enabling EM-accurate component synthesis with real-time inference. The framework successfully generates DRC-aware GDSII layouts for RF circuits, repr
The hypothalamus is most often associated with innate behaviors such as is hunger, thirst and sex. While the expression of these behaviors important for survival of the individual or the species is nested within the hypothalamus, the desire (i.e., motivation) for them is centered within the mesolimbic reward circuitry. In this review, we will use female sexual behavior as a model to examine the interaction of these circuits. We will examine the evidence for a hypothalamic circuit that regulates consummatory aspects of reproductive behavior, i.e. lordosis behavior, a measure of sexual receptivity that involves estradiol membrane-initiated signaling in the arcuate nucleus (ARH), activating β-endorphin projections to the medial preoptic nucleus (MPN), which in turn modulate ventromedial hypothalamic nucleus (VMH) activity – the common output from the hypothalamus. Estradiol modulates not only a series of neuropeptides, transmitters and receptors but induces dendritic spines that are for estrogenic induction of lordosis behavior. Simultaneously, in the nucleus accumbens of the mesolimbic system, the mating experience produces long term changes in dopamine signaling and structure. Sexual experience sensitizes the response of nucleus accumbens neurons to dopamine signaling through the induction of a long lasting early immediate gene. While estrogen alone increases spines in the ARH, sexual experience increases dendritic spine density in the nucleus accumbens. These two circuits appear to converge onto the medial preoptic area where there is a reciprocal influence of motivational circuits on consummatory behavior and vice versa. While it has not been formally demonstrated in the human, such circuitry is generally highly conserved and thus, understanding the anatomy, neurochemistry and physiology can provide useful insight into the motivation for sexual behavior and other innate behaviors in humans.
Ambitious projects aim to record the activity of ever larger and denser neuronal populations in vivo. Correlations in neural activity measured in such recordings can reveal important aspects of neural circuit organization. However, estimating and interpreting large correlation matrices is statistically challenging. Estimation can be improved by regularization, i.e. by imposing a structure on the estimate. The amount of improvement depends on how closely the assumed structure represents dependencies in the data. Therefore, the selection of the most efficient correlation matrix estimator for a given neural circuit must be determined empirically. Importantly, the identity and structure of the most efficient estimator informs about the types of dominant dependencies governing the system. We sought statistically efficient estimators of neural correlation matrices in recordings from large, dense groups of cortical neurons. Using fast 3D random-access laser scanning microscopy of calcium signals, we recorded the activity of nearly every neuron in volumes 200 μm wide and 100 μm deep (150-350 cells) in mouse visual cortex. We hypothesized that in these densely sampled recordings, the correlation matrix should be best modeled as the combination of a sparse graph of pairwise partial correlations representing local interactions and a low-rank component representing common fluctuations and external inputs. Indeed, in cross-validation tests, the covariance matrix estimator with this structure consistently outperformed other regularized estimators. The sparse component of the estimate defined a graph of interactions. These interactions reflected the physical distances and orientation tuning properties of cells: The density of positive 'excitatory' interactions decreased rapidly with geometric distances and with differences in orientation preference whereas negative 'inhibitory' interactions were less selective. Because of its superior performance, this 'sparse+latent' estimator likely provides a more physiologically relevant representation of the functional connectivity in densely sampled recordings than the sample correlation matrix.
The cerebellum regulates complex movements and is also implicated in cognitive tasks, and cerebellar dysfunction is consequently associated not only with movement disorders, but also with conditions like autism and dyslexia. How information is encoded by specific cerebellar firing patterns remains debated, however. A central question is how the cerebellar cortex transmits its integrated output to the cerebellar nuclei via GABAergic synapses from Purkinje neurons. Possible answers come from accumulating evidence that subsets of Purkinje cells synchronize their firing during behaviors that require the cerebellum. Consistent with models predicting that coherent activity of inhibitory networks has the capacity to dictate firing patterns of target neurons, recent experimental work supports the idea that inhibitory synchrony may regulate the response of cerebellar nuclear cells to Purkinje inputs, owing to the interplay between unusually fast inhibitory synaptic responses and high rates of intrinsic activity. Data from multiple laboratories lead to a working hypothesis that synchronous inhibitory input from Purkinje cells can set the timing and rate of action potentials produced by cerebellar nuclear cells, thereby relaying information out of the cerebellum. If so, then changing spatiotemporal patterns of Purkinje activity would allow different subsets of inhibitory neurons to control cerebellar output at different times. Here we explore the evidence for and against the idea that a synchrony code defines, at least in part, the input-output function between the cerebellar cortex and nuclei. We consider the literature on the existence of simple spike synchrony, convergence of Purkinje neurons onto nuclear neurons, and intrinsic properties of nuclear neurons that contribute to responses to inhibition. Finally, we discuss factors that may disrupt or modulate a synchrony code and describe the potential contributions of inhibitory synchrony to other motor circuits.
Recent developments in genetically encoded indicators of neural activity (GINAs) have greatly advanced the field of systems neuroscience. As they are encoded by DNA, GINAs can be targeted to genetically defined cellular populations. Combined with fluorescence microscopy, most notably multi-photon imaging, GINAs allow chronic simultaneous optical recordings from large populations of neurons or glial cells in awake, behaving mammals, particularly rodents. This large-scale recording of neural activity at multiple temporal and spatial scales has greatly advanced our understanding of the dynamics of neural circuitry underlying behavior-a critical first step toward understanding the complexities of brain function, such as sensorimotor integration and learning. Here, we summarize the recent development and applications of the major classes of GINAs. In particular, we take an in-depth look at the design of available GINA families with a particular focus on genetically encoded calcium indicators (GCaMPs), sensors probing synaptic activity, and genetically encoded voltage indicators. Using the family of the GCaMP as an example, we review established sensor optimization pipelines. We also discuss practical considerations for end users of GINAs about experimental methods including approaches for gene delivery, imaging system requirements, and data analysis techniques. With the growing toolbox of GINAs and with new microscopy techniques pushing beyond their current limits, the age of light can finally achieve the goal of broad and dense sampling of neuronal activity across time and brain structures to obtain a dynamic picture of brain function.
Exposure to adverse experiences in early-life is implicated in the later vulnerability to development of psychiatric disorders, including anxiety and affective disorders in humans. Adverse early-life experiences likely impart their long-term consequences on mental health by disrupting the normal development of neural systems involved in stress responses, emotional behavior and emotional states. Neural systems utilizing the neurotransmitters serotonin, dopamine and the neuropeptide corticotropin-releasing factor (CRF) are implicated in mediating emotive behaviors, and dysfunction of these neurochemical systems is associated with mood/anxiety disorders. These neural systems continue maturing until early or mid-adolescence in humans, thus alterations to their development are likely to contribute to the long-term consequences of adverse early-life experiences. A large body of literature suggests that post-weaning isolation rearing of rodents models the behavioral consequences of adverse early-life experiences in humans. Overall, the majority findings suggest that post-weaning social isolation that encompasses pre-adolescence produces long-lasting alterations to anxiety behavior, while measures of monoaminergic activity in various limbic regions during social isolation suggest alterations to dopamine and serotonin systems. The goal of this review is to evaluate and integrate findings from post-weaning social isolation studies specifically related to altered fear and anxiety behaviors and associated changes in neuroendocrine function and the activity of monoaminergic systems.
Koopman operator theory, a powerful framework for discovering the underlying dynamics of nonlinear dynamical systems, was recently shown to be intimately connected with neural network training. In this work, we take the first steps in making use of this connection. As Koopman operator theory is a linear theory, a successful implementation of it in evolving network weights and biases offers the promise of accelerated training, especially in the context of deep networks, where optimization is inherently a non-convex problem. We show that Koopman operator theoretic methods allow for accurate predictions of weights and biases of feedforward, fully connected deep networks over a non-trivial range of training time. During this window, we find that our approach is >10x faster than various gradient descent based methods (e.g. Adam, Adadelta, Adagrad), in line with our complexity analysis. We end by highlighting open questions in this exciting intersection between dynamical systems and neural network theory. We highlight additional methods by which our results could be expanded to broader classes of networks and larger training intervals, which shall be the focus of future work.
Random Neural Networks (RNNs) are a class of Neural Networks (NNs) that can also be seen as a specific type of queuing network. They have been successfully used in several domains during the last 25 years, as queuing networks to analyze the performance of resource sharing in many engineering areas, as learning tools and in combinatorial optimization, where they are seen as neural systems, and also as models of neurological aspects of living beings. In this article we focus on their learning capabilities, and more specifically, we present a practical guide for using the RNN to solve supervised learning problems. We give a general description of these models using almost indistinctly the terminology of Queuing Theory and the neural one. We present the standard learning procedures used by RNNs, adapted from similar well-established improvements in the standard NN field. We describe in particular a set of learning algorithms covering techniques based on the use of first order and, then, of second order derivatives. We also discuss some issues related to these objects and present new perspectives about their use in supervised learning problems. The tutorial describes their most relevant
Recently, dropout has seen increasing use in deep learning. For deep convolutional neural networks, dropout is known to work well in fully-connected layers. However, its effect in convolutional and pooling layers is still not clear. This paper demonstrates that max-pooling dropout is equivalent to randomly picking activation based on a multinomial distribution at training time. In light of this insight, we advocate employing our proposed probabilistic weighted pooling, instead of commonly used max-pooling, to act as model averaging at test time. Empirical evidence validates the superiority of probabilistic weighted pooling. We also empirically show that the effect of convolutional dropout is not trivial, despite the dramatically reduced possibility of over-fitting due to the convolutional architecture. Elaborately designing dropout training simultaneously in max-pooling and fully-connected layers, we achieve state-of-the-art performance on MNIST, and very competitive results on CIFAR-10 and CIFAR-100, relative to other approaches without data augmentation. Finally, we compare max-pooling dropout and stochastic pooling, both of which introduce stochasticity based on multinomial dist
Providing a model that achieves a strong predictive performance and is simultaneously interpretable by humans is one of the most difficult challenges in machine learning research due to the conflicting nature of these two objectives. To address this challenge, we propose a modification of the radial basis function neural network model by equipping its Gaussian kernel with a learnable precision matrix. We show that precious information is contained in the spectrum of the precision matrix that can be extracted once the training of the model is completed. In particular, the eigenvectors explain the directions of maximum sensitivity of the model revealing the active subspace and suggesting potential applications for supervised dimensionality reduction. At the same time, the eigenvectors highlight the relationship in terms of absolute variation between the input and the latent variables, thereby allowing us to extract a ranking of the input variables based on their importance to the prediction task enhancing the model interpretability. We conducted numerical experiments for regression, classification, and feature selection tasks, comparing our model against popular machine learning mode
People often make decisions under aversive conditions such as acute stress. Yet, less is known about the process in which acute stress can influence decision-making. A growing body of research has established that reward-related information associated with the outcomes of decisions exerts a powerful influence over the choices people make and that an extensive network of brain regions, prominently featuring the striatum, is involved in the processing of this reward-related information. Thus, an important step in research on the nature of acute stress' influence over decision-making is to examine how it may modulate responses to rewards and punishments within reward processing neural circuitry. In the current experiment, we employed a simple reward processing paradigm - where participants received monetary rewards and punishments - known to evoke robust striatal responses. Immediately prior to performing each of two task runs, participants were exposed to acute stress (i.e., cold pressor) or a no stress control procedure in a between-subjects fashion. No stress group participants exhibited a pattern of activity within the dorsal striatum and orbitofrontal cortex (OFC) consistent with past research on outcome processing - specifically, differential responses for monetary rewards over punishments. In contrast, acute stress group participants' dorsal striatum and OFC demonstrated decreased sensitivity to monetary outcomes and a lack of differential activity. These findings provide insight into how neural circuits may process rewards and punishments associated with simple decisions under acutely stressful conditions.
A minimal synaptic architecture is proposed for how the brain might perform path integration by computing the next internal representation of self-location from the current representation and from the perceived velocity of motion. In the model, a place-cell assembly called a "chart" contains a two-dimensional attractor set called an "attractor map" that can be used to represent coordinates in any arbitrary environment, once associative binding has occurred between chart locations and sensory inputs. In hippocampus, there are different spatial relations among place fields in different environments and behavioral contexts. Thus, the same units may participate in many charts, and it is shown that the number of uncorrelated charts that can be encoded in the same recurrent network is potentially quite large. According to this theory, the firing of a given place cell is primarily a cooperative effect of the activity of its neighbors on the currently active chart. Therefore, it is not particularly useful to think of place cells as encoding any particular external object or event. Because of its recurrent connections, hippocampal field CA3 is proposed as a possible location for this "multichart" architecture; however, other implementations in anatomy would not invalidate the main concepts. The model is implemented numerically both as a network of integrate-and-fire units and as a "macroscopic" (with respect to the space of states) description of the system, based on a continuous approximation defined by a system of stochastic differential equations. It provides an explanation for a number of hitherto perplexing observations on hippocampal place fields, including doubling, vanishing, reshaping in distorted environments, acquiring directionality in a two-goal shuttling task, rapid formation in a novel environment, and slow rotation after disorientation. The model makes several new predictions about the expected properties of hippocampal place cells and other cells of the proposed network.
Children seem to acquire new know-how in a continuous and open-ended manner. In this paper, we hypothesize that an intrinsic motivation to progress in learning is at the origins of the remarkable structure of children's developmental trajectories. In this view, children engage in exploratory and playful activities for their own sake, not as steps toward other extrinsic goals. The central hypothesis of this paper is that intrinsically motivating activities correspond to expected decrease in prediction error. This motivation system pushes the infant to avoid both predictable and unpredictable situations in order to focus on the ones that are expected to maximize progress in learning. Based on a computational model and a series of robotic experiments, we show how this principle can lead to organized sequences of behavior of increasing complexity characteristic of several behavioral and developmental patterns observed in humans. We then discuss the putative circuitry underlying such an intrinsic motivation system in the brain and formulate two novel hypotheses. The first one is that tonic dopamine acts as a learning progress signal. The second is that this progress signal is directly computed through a hierarchy of microcortical circuits that act both as prediction and metaprediction systems.
Most applications of machine learning to classification assume a closed set of balanced classes. This is at odds with the real world, where class occurrence statistics often follow a long-tailed power-law distribution and it is unlikely that all classes are seen in a single sample. Nonparametric Bayesian models naturally capture this phenomenon, but have significant practical barriers to widespread adoption, namely implementation complexity and computational inefficiency. To address this, we present a method for extracting the inductive bias from a nonparametric Bayesian model and transferring it to an artificial neural network. By simulating data with a nonparametric Bayesian prior, we can metalearn a sequence model that performs inference over an unlimited set of classes. After training, this "neural circuit" has distilled the corresponding inductive bias and can successfully perform sequential inference over an open set of classes. Our experimental results show that the metalearned neural circuit achieves comparable or better performance than particle filter-based methods for inference in these models while being faster and simpler to use than methods that explicitly incorporate