The energy paradigm, exemplified by Hopfield networks, offers a principled framework for memory in neural systems by interpreting dynamics as descent on an energy surface. While powerful for static associative memories, it falls short in modeling sequential memory, where transitions between memories are essential. We introduce the Exponential Dynamic Energy Network (EDEN), a novel architecture that extends the energy paradigm to temporal domains by evolving the energy function over multiple timescales. EDEN combines a static high-capacity energy network with a slow, asymmetrically interacting modulatory population, enabling robust and controlled memory transitions. We formally derive short-timescale energy functions that govern local dynamics and use them to analytically compute memory escape times, revealing a phase transition between static and dynamic regimes. The analysis of capacity, defined as the number of memories that can be stored with minimal error rate as a function of the dimensions of the state space (number of feature neurons), for EDEN shows that it achieves exponential sequence memory capacity $\mathcal{O}(\gamma^N)$, outperforming the linear capacity $\mathcal{O}(N)$ of conventional models. Furthermore, EDEN's dynamics resemble the activity of time and ramping cells observed in the human brain during episodic memory tasks, grounding its biological relevance. By unifying static and sequential memory within a dynamic energy framework, EDEN offers a scalable and interpretable model for high-capacity temporal memory in both artificial and biological systems.

Recent work has revealed that state space models (SSMs), while efficient for long-sequence processing, are fundamentally limited in their ability to represent formal languages—particularly due to time-invariant and real-valued recurrence structures. In this work, we draw inspiration from adaptive and structured dynamics observed in biological neural systems and introduce the Adaptive Unitary State Space Model (AUSSM): a novel class of SSMs that leverages skew-symmetric, input-dependent recurrence to achieve unitary evolution and high expressive power. Using algebraic automata theory, we prove that AUSSM can perform modulo counting and simulate solvable group automata at precision logarithmically bounded in the input length, enabling SSMs to model a broad class of regular languages out of reach for other SSM architectures. To overcome the practical inefficiencies of adaptive recurrence, we develop a separable convolution formulation and a CUDA implementation that enables scalable parallel training. Empirically, we show that AUSSM and its hybrid variant—interleaved with Mamba—outperform prior SSMs on formal algorithmic tasks such as parity and modular arithmetic, and achieve competent performance on real-world long time-series classification benchmarks. Our results demonstrate that adaptive unitary recurrence provides a powerful and efficient inductive bias for both symbolic and continuous sequence modeling. The code is available at https://github.com/arjunkaruvally/AUSSM

Human sensorimotor control is remarkably fast and accurate at the system level despite severe speed-accuracy trade-offs at the component level. The discrepancy between the contrasting speed-accuracy trade-offs at these two levels is a paradox. Meanwhile, speed accuracy trade-offs, heterogeneity, and layered architectures are ubiquitous in nerves, skeletons, and muscles, but they have only been studied in isolation using domain-specific models. In this article, we develop a mechanistic model for how component speed-accuracy trade-offs constrain sensorimotor control that is consistent with Fitts' law for reaching. The model suggests that diversity among components deconstrains the limitations of individual components in sensorimotor control. Such diversity-enabled sweet spots (DESSs) are ubiquitous in nature, explaining why large heterogeneities exist in the components of biological systems and how natural selection routinely evolves systems with fast and accurate responses using imperfect components.

Traveling waves of neural activity have been observed throughout the brain at a diversity of regions and scales; however, their precise computational role is still debated. One physically inspired hypothesis suggests that the cortical sheet may act like a wave-propagating system capable of invertibly storing a short-term memory of sequential stimuli through induced waves traveling across the cortical surface, and indeed many experimental results from neuroscience correlate wave activity with memory tasks. To date, however, the computational implications of this idea have remained hypothetical due to the lack of a simple recurrent neural network architecture capable of exhibiting such waves. In this work, we introduce a model to fill this gap, which we denote the Wave-RNN (wRNN), and demonstrate how such an architecture indeed efficiently encodes the recent past through a suite of synthetic memory tasks where wRNNs learn faster and reach significantly lower error than wave-free counterparts. We further explore the implications of this memory storage system on more complex sequence modeling tasks such as sequential image classification and find that wave-based models not only again outperform comparable wave-free RNNs while using significantly fewer parameters, but additionally perform comparably to more complex gated architectures such as LSTMs and GRUs.

Traveling waves are a fundamental phenomenon in the brain, playing a crucial role in short-term information storage. In this study, we leverage the concept of traveling wave dynamics within a neural lattice to formulate a theoretical model of neural working memory in Recurrent Neural Networks (RNNs), study its properties, and its real world implications in AI. The proposed model diverges from traditional approaches, which assume information storage in static, register-like locations updated by interference. Instead, the model stores data as waves that is updated by the wave's boundary conditions. We rigorously examine the model's capabilities in representing and learning state histories, which are vital for learning history-dependent dynamical systems. The findings reveal that the model reliably stores external information and enhances the learning process by addressing the diminishing gradient problem of RNNs. To understand the model's real-world applicability, we explore two cases: linear boundary condition and non-linear, self-attention-driven boundary condition. The experiments reveal that the linear scenario is effectively *learned* by RNNs through backpropagation when modeling history-dependent dynamical systems. Conversely, the non-linear scenario parallels an attention-only transformer. Collectively, our findings suggest the broader relevance of traveling waves in AI and its potential in advancing neural network architectures.

The current reinforcement learning framework focuses exclusively on performance, often at the expense of efficiency. In contrast, biological control achieves remarkable performance while also optimizing computational energy expenditure and decision frequency. We propose a decision-bounded Markov decision process (DB-MDP) that constrains the number of decisions and computational energy available to agents in reinforcement learning environments. Our experiments demonstrate that existing reinforcement learning algorithms struggle within this framework, leading to either failure or suboptimal performance. To address this, we introduce a biologically inspired, temporally layered architecture (TLA), enabling agents to manage computational costs through two layers with distinct timescales and energy requirements. TLA achieves optimal performance in decision-bounded environments and in continuous control environments, matching state-of-the-art performance while using a fraction of the computing cost. Compared to current reinforcement learning algorithms that solely prioritize performance, our approach significantly lowers computational energy expenditure while maintaining performance. These findings establish a benchmark and pave the way for future research on energy and time-aware control.

Variation in the strength of synapses can be quantified by measuring the anatomical properties of synapses. Quantifying precision of synaptic plasticity is fundamental to understanding information storage and retrieval in neural circuits. Synapses from the same axon onto the same dendrite have a common history of coactivation, making them ideal candidates for determining the precision of synaptic plasticity based on the similarity of their physical dimensions. Here, the precision and amount of information stored in synapse dimensions were quantified with Shannon information theory, expanding prior analysis that used signal detection theory (Bartol et al., 2015). The two methods were compared using dendritic spine head volumes in the middle of the stratum radiatum of hippocampal area CA1 as well-defined measures of synaptic strength. Information theory delineated the number of distinguishable synaptic strengths based on nonoverlapping bins of dendritic spine head volumes. Shannon entropy was applied to measure synaptic information storage capacity (SISC) and resulted in a lower bound of 4.1 bits and upper bound of 4.59 bits of information based on 24 distinguishable sizes. We further compared the distribution of distinguishable sizes and a uniform distribution using Kullback-Leibler divergence and discovered that there was a nearly uniform distribution of spine head volumes across the sizes, suggesting optimal use of the distinguishable values. Thus, SISC provides a new analytical measure that can be generalized to probe synaptic strengths and capacity for plasticity in different brain regions of different species and among animals raised in different conditions or during learning. How brain diseases and disorders affect the precision of synaptic plasticity can also be probed.

Biochemical signaling pathways in living cells are often highly organized into spatially segregated volumes, membranes, scaffolds, subcellular compartments, and organelles comprising small numbers of interacting molecules. At this level of granularity stochastic behavior dominates, well-mixed continuum approximations based on concentrations break down and a particle-based approach is more accurate and more efficient. We describe and validate a new version of the open-source MCell simulation program (MCell4), which supports generalized 3D Monte Carlo modeling of diffusion and chemical reaction of discrete molecules and macromolecular complexes in solution, on surfaces representing membranes, and combinations thereof. The main improvements in MCell4 compared to the previous versions, MCell3 and MCell3-R, include a Python interface and native BioNetGen reaction language (BNGL) support. MCell4's Python interface opens up completely new possibilities for interfacing with external simulators to allow creation of sophisticated event-driven multiscale/multiphysics simulations. The native BNGL support, implemented through a new open-source library libBNG (also introduced in this paper), provides the capability to run a given BNGL model spatially resolved in MCell4 and, with appropriate simplifying assumptions, also in the BioNetGen simulation environment, greatly accelerating and simplifying model validation and comparison.

The state-of-the-art memory model is the General Associative Memory Model, a generalization of the classical Hopfield network. Like its ancestor, the general associative memory has a well-defined state-dependant energy surface, and its memories correlate with its fixed points. This is unlike human memories, which are commonly sequential rather than separated fixed points. In this paper, we introduce a class of General Sequential Episodic Memory Models (GSEMM) that, in the adiabatic limit, exhibit a dynamic energy surface, leading to a series of meta-stable states capable of encoding memory sequences. A multiple-timescale architecture enables the dynamic nature of the energy surface with newly introduced asymmetric synapses and signal propagation delays. We demonstrate its dense capacity under polynomial activation functions. GSEMM combines separate memories, short and long sequential episodic memories, under a unified theoretical framework, demonstrating how energy-based memory modeling can provide richer, human-like episodes.

Large language models (LLMs) have been transformative. They are pretrained foundational models that are self-supervised and can be adapted with fine-tuning to a wide range of natural language tasks, each of which previously would have required a separate network model. This is one step closer to the extraordinary versatility of human language. GPT-3 and, more recently, LaMDA, both of them LLMs, can carry on dialogs with humans on many topics after minimal priming with a few examples. However, there has been a wide range of reactions and debate on whether these LLMs understand what they are saying or exhibit signs of intelligence. This high variance is exhibited in three interviews with LLMs reaching wildly different conclusions. A new possibility was uncovered that could explain this divergence. What appears to be intelligence in LLMs may in fact be a mirror that reflects the intelligence of the interviewer, a remarkable twist that could be considered a reverse Turing test. If so, then by studying interviews, we may be learning more about the intelligence and beliefs of the interviewer than the intelligence of the LLMs. As LLMs become more capable, they may transform the way we interact with machines and how they interact with each other. Increasingly, LLMs are being coupled with sensorimotor devices. LLMs can talk the talk, but can they walk the walk? A road map for achieving artificial general autonomy is outlined with seven major improvements inspired by brain systems and how LLMs could in turn be used to uncover new insights into brain function.

Animal models are used to understand principles of human biology. Within cognitive neuroscience, non-human primates are considered the premier model for studying decision-making behaviors in which direct manipulation experiments are still possible. Some prominent studies have brought to light major discrepancies between monkey and human cognition, highlighting problems with unverified extrapolation from monkey to human. Here, we use a parallel model system-artificial neural networks (ANNs)-to investigate a well-established discrepancy identified between monkeys and humans with a working memory task, in which monkeys appear to use a recency-based strategy while humans use a target-selective strategy. We find that ANNs trained on the same task exhibit a progression of behavior from random behavior (untrained) to recency-like behavior (partially trained) and finally to selective behavior (further trained), suggesting monkeys and humans may occupy different points in the same overall learning progression. Surprisingly, what appears to be recency-like behavior in the ANN, is in fact an emergent non-recency-based property of the organization of the neural network's state space during its development through training. We find that explicit encouragement of recency behavior during training has a dual effect, not only causing an accentuated recency-like behavior, but also speeding up the learning process altogether, resulting in an efficient shaping mechanism to achieve the optimal strategy. Our results suggest a new explanation for the discrepency observed between monkeys and humans and reveal that what can appear to be a recency-based strategy in some cases may not be recency at all.

Chemical synapses exhibit a diverse array of internal mechanisms that affect the dynamics of transmission efficacy. Many of these processes, such as release of neurotransmitter and vesicle recycling, depend strongly on activity-dependent influx and accumulation of Ca2+. To model how each of these processes may affect the processing of information in neural circuits, and how their dysfunction may lead to disease states, requires a computationally efficient modelling framework, capable of generating accurate phenomenology without incurring a heavy computational cost per synapse. Constructing a phenomenologically realistic model requires the precise characterization of the timing and probability of neurotransmitter release. Difficulties arise in that functional forms of instantaneous release rate can be difficult to extract from noisy data without running many thousands of trials, and in biophysical synapses, facilitation of per-vesicle release probability is confounded by depletion. To overcome this, we obtained traces of free Ca2+ concentration in response to various action potential stimulus trains from a molecular MCell model of a hippocampal Schaffer collateral axon. Ca2+ sensors were placed at varying distance from a voltage-dependent calcium channel (VDCC) cluster, and Ca2+ was buffered by calbindin. Then, using the calcium traces to drive deterministic state vector models of synaptotagmin 1 and 7 (Syt-1/7), which respectively mediate synchronous and asynchronous release in excitatory hippocampal synapses, we obtained high-resolution profiles of instantaneous release rate, to which we applied functional fits. Synchronous vesicle release occurred predominantly within half a micron of the source of spike-evoked Ca2+ influx, while asynchronous release occurred more consistently at all distances. Both fast and slow mechanisms exhibited multi-exponential release rate curves, whose magnitudes decayed exponentially with distance from the Ca2+ source. Profile parameters facilitate on different time scales according to a single, general facilitation function. These functional descriptions lay the groundwork for efficient mesoscale modelling of vesicular release dynamics.

In recent years, the use of bio-sensing signals such as electroencephalogram (EEG), electrocardiogram (ECG), etc. have garnered interest towards applications in affective computing. The parallel trend of deep-learning has led to a huge leap in performance towards solving various vision-based research problems such as object detection. Yet, these advances in deep-learning have not adequately translated into bio-sensing research. This work applies novel deep-learning-based methods to various bio-sensing and video data of four publicly available multi-modal emotion datasets. For each dataset, we first individually evaluate the emotion-classification performance obtained by each modality. We then evaluate the performance obtained by fusing the features from these modalities. We show that our algorithms outperform the results reported by other studies for emotion/valence/arousal/liking classification on DEAP and MAHNOB-HCI datasets and set up benchmarks for the newer AMIGOS and DREAMER datasets. We also evaluate the performance of our algorithms by combining the datasets and by using transfer learning to show that the proposed method overcomes the inconsistencies between the datasets. Hence, we do a thorough analysis of multi-modal affective data from more than 120 subjects and 2,800 trials. Finally, utilizing a convolution-deconvolution network, we propose a new technique towards identifying salient brain regions corresponding to various affective states.

Replay is the reactivation of one or more neural patterns that are similar to the activation patterns experienced during past waking experiences. Replay was first observed in biological neural networks during sleep, and it is now thought to play a critical role in memory formation, retrieval, and consolidation. Replay-like mechanisms have been incorporated in deep artificial neural networks that learn over time to avoid catastrophic forgetting of previous knowledge. Replay algorithms have been successfully used in a wide range of deep learning methods within supervised, unsupervised, and reinforcement learning paradigms. In this letter, we provide the first comprehensive comparison between replay in the mammalian brain and replay in artificial neural networks. We identify multiple aspects of biological replay that are missing in deep learning systems and hypothesize how they could be used to improve artificial neural networks.

Recurrent neural network (RNN) models trained to perform cognitive tasks are a useful computational tool for understanding how cortical circuits execute complex computations. However, these models are often composed of units that interact with one another using continuous signals and overlook parameters intrinsic to spiking neurons. Here, we developed a method to directly train not only synaptic-related variables but also membrane-related parameters of a spiking RNN model. Training our model on a wide range of cognitive tasks resulted in diverse yet task-specific synaptic and membrane parameters. We also show that fast membrane time constants and slow synaptic decay dynamics naturally emerge from our model when it is trained on tasks associated with working memory (WM). Further dissecting the optimized parameters revealed that fast membrane properties are important for encoding stimuli, and slow synaptic dynamics are needed for WM maintenance. This approach offers a unique window into how connectivity patterns and intrinsic neuronal properties contribute to complex dynamics in neural populations.

Neuroscience and artificial intelligence (AI) share a long history of collaboration. Advances in neuroscience, alongside huge leaps in computer processing power over the last few decades, have given rise to a new generation of in silico neural networks inspired by the architecture of the brain. These AI systems are now capable of many of the advanced perceptual and cognitive abilities of biological systems, including object recognition and decision making. Moreover, AI is now increasingly being employed as a tool for neuroscience research and is transforming our understanding of brain functions. In particular, deep learning has been used to model how convolutional layers and recurrent connections in the brain's cerebral cortex control important functions, including visual processing, memory, and motor control. Excitingly, the use of neuroscience-inspired AI also holds great promise for understanding how changes in brain networks result in psychopathologies, and could even be utilized in treatment regimes. Here we discuss recent advancements in four areas in which the relationship between neuroscience and AI has led to major advancements in the field; (1) AI models of working memory, (2) AI visual processing, (3) AI analysis of big neuroscience datasets, and (4) computational psychiatry.

Parkinson's disease (PD) is the second most prevalent neurodegenerative disorder in the world. PD is known to lead to marked alterations in cortico-thalamo-basal ganglia activity and subsequent movements, which may provide a biomarker for PD diagnosis. Delay differential analysis (DDA) is a time domain analysis framework based on embedding theory in nonlinear dynamics. An embedding reveals the nonlinear invariant properties of an unknown dynamical system (here the brain) from a single time series electroencephalography (EEG) or behavioral signals. The DDA models serve as a low-dimensional nonlinear functional basis onto which the data are mapped. The combination of behavioral and neurological observations gives rise to a multimodal analysis framework that could improve our understanding and classification of PD using time series data from physical systems. We demonstrate how 750 ms of multimodal data can be used to improve DDA classification performance of PD, over clean EEG or behavioral time series data on their own, in two distinct virtual reach to grasp tasks in an uncertain and dynamic virtual reality environment. Thus, multimodal DDA may provide a tool for aiding the clinician in the diagnosis of PD and bolster classification performance through the combination of a wide array of neural or behavioral signals.

Measuring functional connectivity from fMRI recordings is important in understanding processing in cortical networks. However, because the brain's connection pattern is complex, currently used methods are prone to producing false functional connections. We introduce differential covariance analysis, a new method that uses derivatives of the signal for estimating functional connectivity. We generated neural activities from dynamical causal modeling and a neural network of Hodgkin-Huxley neurons and then converted them to hemodynamic signals using the forward balloon model. The simulated fMRI signals, together with the ground-truth connectivity pattern, were used to benchmark our method with other commonly used methods. Differential covariance achieved better results in complex network simulations. This new method opens an alternative way to estimate functional connectivity.

Calmodulin-dependent kinase II (CaMKII) has long been known to play an important role in learning and memory as well as long term potentiation (LTP). More recently it has been suggested that it might be involved in the time averaging of synaptic signals, which can then lead to the high precision of information stored at a single synapse. However, the role of the scaffolding molecule, neurogranin (Ng), in governing the dynamics of CaMKII is not yet fully understood. In this work, we adopt a rule-based modeling approach through the Monte Carlo method to study the effect of Ca2+ signals on the dynamics of CaMKII phosphorylation in the postsynaptic density (PSD). Calcium surges are observed in synaptic spines during an EPSP and back-propagating action potential due to the opening of NMDA receptors and voltage dependent calcium channels. Using agent-based models, we computationally investigate the dynamics of phosphorylation of CaMKII monomers and dodecameric holoenzymes. The scaffolding molecule, Ng, when present in significant concentration, limits the availability of free calmodulin (CaM), the protein which activates CaMKII in the presence of calcium. We show that Ng plays an important modulatory role in CaMKII phosphorylation following a surge of high calcium concentration. We find a non-intuitive dependence of this effect on CaM concentration that results from the different affinities of CaM for CaMKII depending on the number of calcium ions bound to the former. It has been shown previously that in the absence of phosphatase, CaMKII monomers integrate over Ca2+ signals of certain frequencies through autophosphorylation (Pepke et al, Plos Comp. Bio., 2010). We also study the effect of multiple calcium spikes on CaMKII holoenzyme autophosphorylation, and show that in the presence of phosphatase, CaMKII behaves as a leaky integrator of calcium signals, a result that has been recently observed in vivo. Our models predict that the parameters of this leaky integrator are finely tuned through the interactions of Ng, CaM, CaMKII, and PP1, providing a mechanism to precisely control the sensitivity of synapses to calcium signals. Author Summary not valid for PLOS ONE submissions.

Epilepsy is a neurological disorder characterized by the sudden occurrence of unprovoked seizures. There is extensive evidence of significantly altered brain connectivity during seizure periods in the human brain. Research on analyzing human brain functional connectivity during epileptic seizures has been limited predominantly to the use of the correlation method. However, spurious connectivity can be measured between two brain regions without having direct connection or interaction between them. Correlations can be due to the apparent interactions of the two brain regions resulting from common input from a third region, which may or may not be observed. Hence, researchers have recently proposed a sparse-plus-latent-regularized precision matrix (SLRPM) when there are unobserved or latent regions interacting with the observed regions. The SLRPM method yields partial correlations of the conditional statistics of the observed regions given the latent regions, thus identifying observed regions that are conditionally independent of both the observed and latent regions. We evaluate the performance of the methods using a spring-mass artificial network and assuming that some nodes cannot be observed, thus constituting the latent variables in the example. Several cases have been considered, including both sparse and dense connections, short-range and long-range connections, and a varying number of latent variables. The SLRPM method is then applied to estimate brain connectivity during epileptic seizures from human ECoG recordings. Seventy-four clinical seizures from five patients, all having complex partial epilepsy, were analyzed using SLRPM, and brain connectivity was quantified using modularity index, clustering coefficient, and eigenvector centrality. Furthermore, using a measure of latent inputs estimated by the SLRPM method, it was possible to automatically detect 72 of the 74 seizures with four false positives and find six seizures that were not marked manually.

Ca2+/calmodulin-dependent protein kinase II (CaMKII) accounts for up to 2 percent of all brain protein and is essential to memory function. CaMKII activity is known to regulate dynamic shifts in the size and signaling strength of neuronal connections, a process known as synaptic plasticity. Increasingly, computational models are used to explore synaptic plasticity and the mechanisms regulating CaMKII activity. Conventional modeling approaches may exclude biophysical detail due to the impractical number of state combinations that arise when explicitly monitoring the conformational changes, ligand binding, and phosphorylation events that occur on each of the CaMKII holoenzyme's subunits. To manage the combinatorial explosion without necessitating bias or loss in biological accuracy, we use a specialized syntax in the software MCell to create a rule-based model of a twelve-subunit CaMKII holoenzyme. Here we validate the rule-based model against previous experimental measures of CaMKII activity and investigate molecular mechanisms of CaMKII regulation. Specifically, we explore how Ca2+/CaM-binding may both stabilize CaMKII subunit activation and regulate maintenance of CaMKII autophosphorylation. Noting that Ca2+/CaM and protein phosphatases bind CaMKII at nearby or overlapping sites, we compare model scenarios in which Ca2+/CaM and protein phosphatase do or do not structurally exclude each other's binding to CaMKII. Our results suggest a functional mechanism for the so-called "CaM trapping" phenomenon, wherein Ca2+/CaM may structurally exclude phosphatase binding and thereby prolong CaMKII autophosphorylation. We conclude that structural protection of autophosphorylated CaMKII by Ca2+/CaM may be an important mechanism for regulation of synaptic plasticity.

Most large-scale network models use neurons with static nonlinearities that produce analog output, despite the fact that information processing in the brain is predominantly carried out by dynamic neurons that produce discrete pulses called spikes. Research in spike-based computation has been impeded by the lack of efficient supervised learning algorithm for spiking neural networks. Here, we present a gradient descent method for optimizing spiking network models by introducing a differentiable formulation of spiking dynamics and deriving the exact gradient calculation. For demonstration, we trained recurrent spiking networks on two dynamic tasks: one that requires optimizing fast (~ millisecond) spike-based interactions for efficient encoding of information, and a delayed-memory task over extended duration (~ second). The results show that the gradient descent approach indeed optimizes networks dynamics on the time scale of individual spikes as well as on behavioral time scales. In conclusion, our method yields a general purpose supervised learning algorithm for spiking neural networks, which can facilitate further investigations on spike-based computations.

Sleep spindles are brief oscillatory events during non-rapid eye movement (NREM) sleep. Spindle density and synchronization properties are different in MEG versus EEG recordings in humans and also vary with learning performance, suggesting spindle involvement in memory consolidation. Here, using computational models, we identified network mechanisms that may explain differences in spindle properties across cortical structures. First, we report that differences in spindle occurrence between MEG and EEG data may arise from the contrasting properties of the core and matrix thalamocortical systems. The matrix system, projecting superficially, has wider thalamocortical fanout compared to the core system, which projects to middle layers, and requires the recruitment of a larger population of neurons to initiate a spindle. This property was sufficient to explain lower spindle density and higher spatial synchrony of spindles in the superficial cortical layers, as observed in the EEG signal. In contrast, spindles in the core system occurred more frequently but less synchronously, as observed in the MEG recordings. Furthermore, consistent with human recordings, in the model, spindles occurred independently in the core system but the matrix system spindles commonly co-occurred with core spindles. We also found that the intracortical excitatory connections from layer III/IV to layer V promote spindle propagation from the core to the matrix system, leading to widespread spindle activity. Our study predicts that plasticity of intra- and inter-cortical connectivity can potentially be a mechanism for increased spindle density as has been observed during learning.

The correlation method from brain imaging has been used to estimate functional connectivity in the human brain. However, brain regions might show very high correlation even when the two regions are not directly connected due to the strong interaction of the two regions with common input from a third region. One previously proposed solution to this problem is to use a sparse regularized inverse covariance matrix or precision matrix (SRPM) assuming that the connectivity structure is sparse. This method yields partial correlations to measure strong direct interactions between pairs of regions while simultaneously removing the influence of the rest of the regions, thus identifying regions that are conditionally independent. To test our methods, we first demonstrated conditions under which the SRPM method could indeed find the true physical connection between a pair of nodes for a spring-mass example and an RC circuit example. The recovery of the connectivity structure using the SRPM method can be explained by energy models using the Boltzmann distribution. We then demonstrated the application of the SRPM method for estimating brain connectivity during stage 2 sleep spindles from human electrocorticography (ECoG) recordings using an [Formula: see text] electrode array. The ECoG recordings that we analyzed were from a 32-year-old male patient with long-standing pharmaco-resistant left temporal lobe complex partial epilepsy. Sleep spindles were automatically detected using delay differential analysis and then analyzed with SRPM and the Louvain method for community detection. We found spatially localized brain networks within and between neighboring cortical areas during spindles, in contrast to the case when sleep spindles were not present.

In estimating the frequency spectrum of real-world time series data, we must violate the assumption of infinite-length, orthogonal components in the Fourier basis. While it is widely known that care must be taken with discretely sampled data to avoid aliasing of high frequencies, less attention is given to the influence of low frequencies with period below the sampling time window. Here, we derive an analytic expression for the side-lobe attenuation of signal components in the frequency domain representation. This expression allows us to detail the influence of individual frequency components throughout the spectrum. The first consequence is that the presence of low-frequency components introduces a 1/f[Formula: see text] component across the power spectrum, with a scaling exponent of [Formula: see text]. This scaling artifact could be composed of diffuse low-frequency components, which can render it difficult to detect a priori. Further, treatment of the signal with standard digital signal processing techniques cannot easily remove this scaling component. While several theoretical models have been introduced to explain the ubiquitous 1/f[Formula: see text] scaling component in neuroscientific data, we conjecture here that some experimental observations could be the result of such data analysis procedures.

High-density electrocorticogram (ECoG) electrodes are capable of recording neurophysiological data with high temporal resolution with wide spatial coverage. These recordings are a window to understanding how the human brain processes information and subsequently behaves in healthy and pathologic states. Here, we describe and implement delay differential analysis (DDA) for the characterization of ECoG data obtained from human patients with intractable epilepsy. DDA is a time-domain analysis framework based on embedding theory in nonlinear dynamics that reveals the nonlinear invariant properties of an unknown dynamical system. The DDA embedding serves as a low-dimensional nonlinear dynamical basis onto which the data are mapped. This greatly reduces the risk of overfitting and improves the method's ability to fit classes of data. Since the basis is built on the dynamical structure of the data, preprocessing of the data (e.g., filtering) is not necessary. We performed a large-scale search for a DDA model that best fit ECoG recordings using a genetic algorithm to qualitatively discriminate between different cortical states and epileptic events for a set of 13 patients. A single DDA model with only three polynomial terms was identified. Singular value decomposition across the feature space of the model revealed both global and local dynamics that could differentiate electrographic and electroclinical seizures and provided insights into highly localized seizure onsets and diffuse seizure terminations. Other common ECoG features such as interictal periods, artifacts, and exogenous stimuli were also analyzed with DDA. This novel framework for signal processing of seizure information demonstrates an ability to reveal unique characteristics of the underlying dynamics of the seizure and may be useful in better understanding, detecting, and maybe even predicting seizures.

With our ability to record more neurons simultaneously, making sense of these data is a challenge. Functional connectivity is one popular way to study the relationship of multiple neural signals. Correlation-based methods are a set of currently well-used techniques for functional connectivity estimation. However, due to explaining away and unobserved common inputs (Stevenson, Rebesco, Miller, & Körding, 2008 ), they produce spurious connections. The general linear model (GLM), which models spike trains as Poisson processes (Okatan, Wilson, & Brown, 2005 ; Truccolo, Eden, Fellows, Donoghue, & Brown, 2005 ; Pillow et al., 2008 ), avoids these confounds. We develop here a new class of methods by using differential signals based on simulated intracellular voltage recordings. It is equivalent to a regularized AR(2) model. We also expand the method to simulated local field potential recordings and calcium imaging. In all of our simulated data, the differential covariance-based methods achieved performance better than or similar to the GLM method and required fewer data samples. This new class of methods provides alternative ways to analyze neural signals.

Our understanding of neural population coding has been limited by a lack of analysis methods to characterize spiking data from large populations. The biggest challenge comes from the fact that the number of possible network activity patterns scales exponentially with the number of neurons recorded ([Formula: see text]). Here we introduce a new statistical method for characterizing neural population activity that requires semi-independent fitting of only as many parameters as the square of the number of neurons, requiring drastically smaller data sets and minimal computation time. The model works by matching the population rate (the number of neurons synchronously active) and the probability that each individual neuron fires given the population rate. We found that this model can accurately fit synthetic data from up to 1000 neurons. We also found that the model could rapidly decode visual stimuli from neural population data from macaque primary visual cortex about 65 ms after stimulus onset. Finally, we used the model to estimate the entropy of neural population activity in developing mouse somatosensory cortex and, surprisingly, found that it first increases, and then decreases during development. This statistical model opens new options for interrogating neural population data and can bolster the use of modern large-scale in vivo Ca[Formula: see text] and voltage imaging tools.

Natural images are scale invariant with structures at all length scales.We formulated a geometric view of scale invariance in natural images using percolation theory, which describes the behavior of connected clusters on graphs.We map images to the percolation model by defining clusters on a binary representation for images. We show that critical percolating structures emerge in natural images and study their scaling properties by identifying fractal dimensions and exponents for the scale-invariant distributions of clusters. This formulation leads to a method for identifying clusters in images from underlying structures as a starting point for image segmentation.

We propose a time-domain approach to detect frequencies, frequency couplings, and phases using nonlinear correlation functions. For frequency analysis, this approach is a multivariate extension of discrete Fourier transform, and for higher-order spectra, it is a linear and multivariate alternative to multidimensional fast Fourier transform of multidimensional correlations. This method can be applied to short and sparse time series and can be extended to cross-trial and cross-channel spectra (CTS) for electroencephalography data where multiple short data segments from multiple trials of the same experiment are available. There are two versions of CTS. The first one assumes some phase coherency across the trials, while the second one is independent of phase coherency. We demonstrate that the phase-dependent version is more consistent with event-related spectral perturbation analysis and traditional Morlet wavelet analysis. We show that CTS can be applied to short data windows and yields higher temporal resolution than traditional Morlet wavelet analysis. Furthermore, the CTS can be used to reconstruct the event-related potential using all linear components of the CTS.

Nonlinear dynamical system analysis based on embedding theory has been used for modeling and prediction, but it also has applications to signal detection and classification of time series. An embedding creates a multidimensional geometrical object from a single time series. Traditionally either delay or derivative embeddings have been used. The delay embedding is composed of delayed versions of the signal, and the derivative embedding is composed of successive derivatives of the signal. The delay embedding has been extended to nonuniform embeddings to take multiple timescales into account. Both embeddings provide information on the underlying dynamical system without having direct access to all the system variables. Delay differential analysis is based on functional embeddings, a combination of the derivative embedding with nonuniform delay embeddings. Small delay differential equation (DDE) models that best represent relevant dynamic features of time series data are selected from a pool of candidate models for detection or classification. We show that the properties of DDEs support spectral analysis in the time domain where nonlinear correlation functions are used to detect frequencies, frequency and phase couplings, and bispectra. These can be efficiently computed with short time windows and are robust to noise. For frequency analysis, this framework is a multivariate extension of discrete Fourier transform (DFT), and for higher-order spectra, it is a linear and multivariate alternative to multidimensional fast Fourier transform of multidimensional correlations. This method can be applied to short or sparse time series and can be extended to cross-trial and cross-channel spectra if multiple short data segments of the same experiment are available. Together, this time-domain toolbox provides higher temporal resolution, increased frequency and phase coupling information, and it allows an easy and straightforward implementation of higher-order spectra across time compared with frequency-based methods such as the DFT and cross-spectral analysis.

Inhibition-stabilized networks (ISNs) are neural architectures with strong positive feedback among pyramidal neurons balanced by strong negative feedback from inhibitory interneurons, a circuit element found in the hippocampus and the primary visual cortex. In their working regime, ISNs produce damped oscillations in the [Formula: see text]-range in response to inputs to the inhibitory population. In order to understand the properties of interconnected ISNs, we investigated periodic forcing of ISNs. We show that ISNs can be excited over a range of frequencies and derive properties of the resonance peaks. In particular, we studied the phase-locked solutions, the torus solutions, and the resonance peaks. Periodically forced ISNs respond with (possibly multistable) phase-locked activity, whereas networks with sustained intrinsic oscillations respond more dynamically to periodic inputs with tori. Hence, the dynamics are surprisingly rich, and phase effects alone do not adequately describe the network response. This strengthens the importance of phase-amplitude coupling as opposed to phase-phase coupling in providing multiple frequencies for multiplexing and routing information.

Data sets with high dimensionality such as natural images, speech, and text have been analyzed with methods from condensed matter physics. Here we compare recent approaches taken to relate the scale invariance of natural images to critical phenomena. We also examine the method of studying high-dimensional data through specific heat curves by applying the analysis to noncritical systems: 1D samples taken from natural images and 2D binary pink noise. Through these examples, we concluded that due to small sample sizes, specific heat is not a reliable measure for gauging whether high-dimensional data are critical. We argue that identifying order parameters and universality classes is a more reliable way to identify criticality in high-dimensional data.

Understanding the anatomical and functional architecture of the brain is essential for designing neurally inspired intelligent systems. Theoretical and empirical studies suggest a role for narrowband oscillations in shaping the functional architecture of the brain through their role in coding and communication of information. Such oscillations are ubiquitous signals in the electrical activity recorded from the brain. In the cortex, oscillations detected in the gamma range (30-80 Hz) are modulated by behavioral states and sensory features in complex ways. How is this regulation achieved? Although several underlying principles for the genesis of these oscillations have been proposed, a unifying account for their regulation has remained elusive. In a network of excitatory and inhibitory neurons operating in an inhibition-stabilized regime, we show that strongly superlinear responses of inhibitory neurons facilitate bidirectional regulation of oscillation frequency and power. In such a network, the balance of drives to the excitatory and inhibitory populations determines how the power and frequency of oscillations are modulated. The model accounts for the puzzling increase in their frequency with the salience of visual stimuli, and a decrease with their size. Oscillations in our model grow stronger as the mean firing level is reduced, accounting for the size dependence of visually evoked gamma rhythms, and suggesting a role for oscillations in improving the signal-to-noise ratio (SNR) of signals in the brain. Empirically testing such predictions is still challenging, and implementing the proposed coding and communication strategies in neuromorphic systems could assist in our understanding of the biological system.

This special issue focuses on elucidating computational neuroscience: an interdisciplinary field of scientific research in which one of the primary goals is to understand how electronic activity in brain cells and networks enables biological intelligence. The objective is to provide a selection of papers that expose and review research efforts in aspects of computational neuroscience that demonstrate its rapidly growing intersection with electrical, electronic and computer engineering, and the prospects for interaction in the near and long-term future.

Human performance approaches that of an ideal observer and optimal actor in some perceptual and motor tasks. These optimal abilities depend on the capacity of the cerebral cortex to store an immense amount of information and to flexibly make rapid decisions. However, behavior only approaches these limits after a long period of learning while the cerebral cortex interacts with the basal ganglia, an ancient part of the vertebrate brain that is responsible for learning sequences of actions directed toward achieving goals. Progress has been made in understanding the algorithms used by the brain during reinforcement learning, which is an online approximation of dynamic programming. Humans also make plans that depend on past experience by simulating different scenarios, which is called prospective optimization. The same brain structures in the cortex and basal ganglia that are active online during optimal behavior are also active offline during prospective optimization. The emergence of general principles and algorithms for goal-directed behavior has consequences for the development of autonomous devices in engineering applications.

Sleep spindles and K-complexes (KCs) define stage 2 NREM sleep (N2) in humans. We recently showed that KCs are isolated downstates characterized by widespread cortical silence. We demonstrate here that KCs can be quasi-synchronous across scalp EEG and across much of the cortex using electrocorticography (ECOG) and localized transcortical recordings (bipolar SEEG). We examine the mechanism of synchronous KC production by creating the first conductance based thalamocortical network model of N2 sleep to generate both spontaneous spindles and KCs. Spontaneous KCs are only observed when the model includes diffuse projections from restricted prefrontal areas to the thalamic reticular nucleus (RE), consistent with recent anatomical findings in rhesus monkeys. Modeled KCs begin with a spontaneous focal depolarization of the prefrontal neurons, followed by depolarization of the RE. Surprisingly, the RE depolarization leads to decreased firing due to disrupted spindling, which in turn is due to depolarization-induced inactivation of the low-threshold Ca2+ current (IT). Further, although the RE inhibits thalamocortical (TC) neurons, decreased RE firing causes decreased TC cell firing, again because of disrupted spindling. The resulting abrupt removal of excitatory input to cortical pyramidal neurons then leads to the downstate. Empirically, KCs may also be evoked by sensory stimuli while maintaining sleep. We reproduce this phenomenon in the model by depolarization of either the RE or the widely-projecting prefrontal neurons. Again, disruption of thalamic spindling plays a key role. Higher levels of RE stimulation also cause downstates, but by directly inhibiting the TC neurons. SEEG recordings from the thalamus and cortex in a single patient demonstrated the model prediction that thalamic spindling significantly decreases before KC onset. In conclusion, we show empirically that KCs can be widespread quasi-synchronous cortical downstates, and demonstrate with the first model of stage 2 NREM sleep a possible mechanism whereby this widespread synchrony may arise.

Perceptual learning has been used to probe the mechanisms of cortical plasticity in the adult brain. Feedback projections are ubiquitous in the cortex, but little is known about their role in cortical plasticity. Here we explore the hypothesis that learning visual orientation discrimination involves learning-dependent plasticity of top-down feedback inputs from higher cortical areas, serving a different function from plasticity due to changes in recurrent connections within a cortical area. In a Hodgkin-Huxley-based spiking neural network model of visual cortex, we show that modulation of feedback inputs to V1 from higher cortical areas results in shunting inhibition in V1 neurons, which changes the response properties of V1 neurons. The orientation selectivity of V1 neurons is enhanced without changing orientation preference, preserving the topographic organizations in V1. These results provide new insights to the mechanisms of plasticity in the adult brain, reconciling apparently inconsistent experiments and providing a new hypothesis for a functional role of the feedback connections.

Episodic-like memory is thought to be supported by attractor dynamics in the hippocampus. A possible neural substrate for this memory mechanism is rate remapping, in which the spatial map of place cells encodes contextual information through firing rate variability. To test whether memories are stored as multimodal attractors in populations of place cells, recent experiments morphed one familiar context into another while observing the responses of CA3 cell ensembles. Average population activity in CA3 was reported to transition gradually rather than abruptly from one familiar context to the next, suggesting a lack of attractive forces associated with the two stored representations. On the other hand, individual CA3 cells showed a mix of gradual and abrupt transitions at different points along the morph sequence, and some displayed hysteresis which is a signature of attractor dynamics. To understand whether these seemingly conflicting results are commensurate with attractor network theory, we developed a neural network model of the CA3 with attractors for both position and discrete contexts. We found that for memories stored in overlapping neural ensembles within a single spatial map, position-dependent context attractors made transitions at different points along the morph sequence. Smooth transition curves arose from averaging across the population, while a heterogeneous set of responses was observed on the single unit level. In contrast, orthogonal memories led to abrupt and coherent transitions on both population and single unit levels as experimentally observed when remapping between two independent spatial maps. Strong recurrent feedback entailed a hysteretic effect on the network which diminished with the amount of overlap in the stored memories. These results suggest that context-dependent memory can be supported by overlapping local attractors within a spatial map of CA3 place cells. Similar mechanisms for context-dependent memory may also be found in other regions of the cerebral cortex.

Multi-state modeling of biomolecules refers to a series of techniques used to represent and compute the behavior of biological molecules or complexes that can adopt a large number of possible functional states. Biological signaling systems often rely on complexes of biological macromolecules that can undergo several functionally significant modifications that are mutually compatible. Thus, they can exist in a very large number of functionally different states. Modeling such multi-state systems poses two problems: the problem of how to describe and specify a multi-state system (the "specification problem") and the problem of how to use a computer to simulate the progress of the system over time (the "computation problem"). To address the specification problem, modelers have in recent years moved away from explicit specification of all possible states and towards rule-based formalisms that allow for implicit model specification, including the κ-calculus, BioNetGen, the Allosteric Network Compiler, and others. To tackle the computation problem, they have turned to particle-based methods that have in many cases proved more computationally efficient than population-based methods based on ordinary differential equations, partial differential equations, or the Gillespie stochastic simulation algorithm. Given current computing technology, particle-based methods are sometimes the only possible option. Particle-based simulators fall into two further categories: nonspatial simulators, such as StochSim, DYNSTOC, RuleMonkey, and the Network-Free Stochastic Simulator (NFSim), and spatial simulators, including Meredys, SRSim, and MCell. Modelers can thus choose from a variety of tools, the best choice depending on the particular problem. Development of faster and more powerful methods is ongoing, promising the ability to simulate ever more complex signaling processes in the future.

The analysis of natural images with independent component analysis (ICA) yields localized bandpass Gabor-type filters similar to receptive fields of simple cells in visual cortex. We applied ICA on a subset of patches called position-centered patches, selected for forming a translation-invariant representation of small patches. The resulting filters were qualitatively different in two respects. One novel feature was the emergence of filters we call double-Gabor filters. In contrast to Gabor functions that are modulated in one direction, double-Gabor filters are sinusoidally modulated in two orthogonal directions. In addition the filters were more extended in space and frequency compared to standard ICA filters and better matched the distribution in experimental recordings from neurons in primary visual cortex. We further found a dual role for double-Gabor filters as edge and texture detectors, which could have engineering applications.

Traumatic brain injury often leads to epileptic seizures. Among other factors, homeostatic synaptic plasticity (HSP) mediates posttraumatic epileptogenesis through unbalanced synaptic scaling, partially compensating for the trauma-incurred loss of neural excitability. HSP is mediated in part by tumor necrosis factor alpha (TNFα), which is released locally from reactive astrocytes early after trauma in response to chronic neuronal inactivity. During this early period, TNFα is likely to be constrained to its glial sources; however, the contribution of glia-mediated spatially localized HSP to post-traumatic epileptogenesis remains poorly understood. We used computational model to investigate the reorganization of collective neural activity early after trauma. Trauma and synaptic scaling transformed asynchronous spiking into paroxysmal discharges. The rate of paroxysms could be reduced by functional segregation of synaptic input into astrocytic microdomains. Thus, we propose that trauma-triggered reactive gliosis could exert both beneficial and deleterious effects on neural activity.

How neurons in the cerebral cortex process and transmit information is a long-standing question in systems neuroscience. To analyze neuronal activity from an information-energy efficiency standpoint, Berger and Levy calculated the maximum Shannon mutual information transfer per unit of energy expenditure of an idealized integrate-and-fire (IIF) neuron whose excitatory synapses all have the same weight. Here, we extend their IIF model to a biophysically more realistic one in which synaptic weights are unequal. Using information theory, random Poisson measures, and the maximum entropy principle, we show that the probability density function (pdf) of interspike interval (ISI) duration induced by the bits per joule (bpj) maximizing pdf fΛ(λ) of the excitatory postsynaptic potential (EPSP) intensity remains equal to the delayed gamma distribution of the IIF model. We then show that, in the case of unequal weights, fΛ(·) satisfies an inhomogeneous Cauchy-Euler equation with variable coefficients for which we provide the general solution form.

When subjects adapt their reaching movements in the setting of a systematic force or visual perturbation, generalization of adaptation can be assessed psychophysically in two ways: by testing untrained locations in the work space at the end of adaptation (slow postadaptation generalization) or by determining the influence of an error on the next trial during adaptation (fast trial-by-trial generalization). These two measures of generalization have been widely used in psychophysical studies, but the reason that they might differ has not been addressed explicitly. Our goal was to develop a computational framework for determining when a two-state model is justified by the data and to explore the implications of these two types of generalization for neural representations of movements. We first investigated, for single-target learning, how well standard statistical model selection procedures can discriminate two-process models from single-process models when learning and retention coefficients were systematically varied. We then built a two-state model for multitarget learning and showed that if an adaptation process is indeed two-rate, then the postadaptation generalization approach primarily probes the slow process, whereas the trial-by-trial generalization approach is most informative about the fast process. The fast process, due to its strong sensitivity to trial error, contributes predominantly to trial-by-trial generalization, whereas the strong retention of the slow system contributes predominantly to postadaptation generalization. Thus, when adaptation can be shown to be two-rate, the two measures of generalization may probe different brain representations of movement direction.

The response of a neuron to repeated somatic fluctuating current injections in vitro can elicit a reliable and precisely timed sequence of action potentials. The set of responses obtained across trials can also be interpreted as the response of an ensemble of similar neurons receiving the same input, with the precise spike times representing synchronous volleys that would be effective in driving postsynaptic neurons. To study the reproducibility of the output spike times for different conditions that might occur in vivo, we somatically injected aperiodic current waveforms into cortical neurons in vitro and systematically varied the amplitude and DC offset of the fluctuations. As the amplitude of the fluctuations was increased, reliability increased and the spike times remained stable over a wide range of values. However, at specific values called bifurcation points, large shifts in the spike times were obtained in response to small changes in the stimulus, resulting in multiple spike patterns that were revealed using an unsupervised classification method. Increasing the DC offset, which mimicked an overall increase in network background activity, also revealed bifurcation points and increased the reliability. Furthermore, the spike times shifted earlier with increasing offset. Although the reliability was reduced at bifurcation points, a theoretical analysis showed that the information about the stimulus time course was increased because each of the spike time patterns contained different information about the input.

The pattern of spikes recorded from place cells in the rodent hippocampus is strongly modulated by both the spatial location in the environment and the theta rhythm. The phases of the spikes in the theta cycle advance during movement through the place field. Recently intracellular recordings from hippocampal neurons (Harvey, Collman, Dombeck, & Tank, 2009 ) showed an increase in the amplitude of membrane potential oscillations inside the place field, which was interpreted as evidence that an intracellular mechanism caused phase precession. Here we show that an existing network model of the hippocampus (Tsodyks, Skaggs, Sejnowski, & McNaughton, 1996 ) can equally reproduce this and other aspects of the intracellular recordings, which suggests that new experiments are needed to distinguish the contributions of intracellular and network mechanisms to phase precession.

Neurons in sensory systems convey information about physical stimuli in their spike trains. In vitro, single neurons respond precisely and reliably to the repeated injection of the same fluctuating current, producing regions of elevated firing rate, termed events. Analysis of these spike trains reveals that multiple distinct spike patterns can be identified as trial-to-trial correlations between spike times (Fellous, Tiesinga, Thomas, & Sejnowski, 2004 ). Finding events in data with realistic spiking statistics is challenging because events belonging to different spike patterns may overlap. We propose a method for finding spiking events that uses contextual information to disambiguate which pattern a trial belongs to. The procedure can be applied to spike trains of the same neuron across multiple trials to detect and separate responses obtained during different brain states. The procedure can also be applied to spike trains from multiple simultaneously recorded neurons in order to identify volleys of near-synchronous activity or to distinguish between excitatory and inhibitory neurons. The procedure was tested using artificial data as well as recordings in vitro in response to fluctuating current waveforms.

Convolutive mixtures of signals, which are common in acoustic environments, can be difficult to separate into their component sources. Here we present a uniform probabilistic framework to separate convolutive mixtures of acoustic signals using independent vector analysis (IVA), which is based on a joint distribution for the frequency components originating from the same source and is capable of preventing permutation disorder. Different gaussian mixture models (GMM) served as source priors, in contrast to the original IVA model, where all sources were modeled by identical multivariate Laplacian distributions. This flexible source prior enabled the IVA model to separate different type of signals. Three classes of models were derived and tested: noiseless IVA, online IVA, and noisy IVA. In the IVA model without sensor noise, the unmixing matrices were efficiently estimated by the expectation maximization (EM) algorithm. An online EM algorithm was derived for the online IVA algorithm to track the movement of the sources and separate them under nonstationary conditions. The noisy IVA model included the sensor noise and combined denoising with separation. An EM algorithm was developed that found the model parameters and separated the sources simultaneously. These algorithms were applied to separate mixtures of speech and music. Performance as measured by the signal-to-interference ratio (SIR) was substantial for all three models.

Perceiving and identifying an object is improved by prior exposure to the object. This perceptual priming phenomenon is accompanied by reduced neural activity. But whether suppression of neuronal activity with priming is responsible for the improvement in perception is unclear. To address this problem, we developed a rate-based network model of visual processing. In the model, decreased neural activity following priming was due to stimulus-specific sharpening of representations taking place in the early visual areas. Representation sharpening led to decreased interference of representations in higher visual areas that facilitated selection of one of the competing representations, thereby improving recognition. The model explained a wide range of psychophysical and physiological data observed in priming experiments, including antipriming phenomena, and predicted two functionally distinct stages of visual processing.

We study local calcium dynamics leading to a vesicle fusion in a stochastic, and spatially explicit, biophysical model of the CA3-CA1 presynaptic bouton. The kinetic model for vesicle release has two calcium sensors, a sensor for fast synchronous release that lasts a few tens of milliseconds and a separate sensor for slow asynchronous release that lasts a few hundred milliseconds. A wide range of data can be accounted for consistently only when a refractory period lasting a few milliseconds between releases is included. The inclusion of a second sensor for asynchronous release with a slow unbinding site, and thereby a long memory, affects short-term plasticity by facilitating release. Our simulations also reveal a third time scale of vesicle release that is correlated with the stimulus and is distinct from the fast and the slow releases. In these detailed Monte Carlo simulations all three time scales of vesicle release are insensitive to the spatial details of the synaptic ultrastructure. Furthermore, our simulations allow us to identify features of synaptic transmission that are universal and those that are modulated by structure.

The sensitivity of a neuron to its input can be modulated in several ways. Changes in the slope of the neuronal input-output curve depend on factors such as shunting inhibition, background noise, frequency-dependent synaptic excitation, and balanced excitation and inhibition. However, in early development GABAergic interneurons are excitatory and other mechanisms such as asynchronous transmitter release might contribute to regulating neuronal sensitivity. We modeled both phasic and asynchronous synaptic transmission in early development to study the impact of activity-dependent noise and short-term plasticity on the synaptic gain. Asynchronous release decreased or increased the gain depending on the membrane conductance. In the high shunt regime, excitatory input due to asynchronous release was divisive, whereas in the low shunt regime it had a nearly multiplicative effect on the firing rate. In addition, sensitivity to correlated inputs was influenced by shunting and asynchronous release in opposite ways. Thus, asynchronous release can regulate the information flow at synapses and its impact can be flexibly modulated by the membrane conductance.

This paper presents a novel probabilistic approach to speech enhancement. Instead of a deterministic logarithmic relationship, we assume a probabilistic relationship between the frequency coefficients and the log-spectra. The speech model in the log-spectral domain is a Gaussian mixture model (GMM). The frequency coefficients obey a zero-mean Gaussian whose covariance equals to the exponential of the log-spectra. This results in a Gaussian scale mixture model (GSMM) for the speech signal in the frequency domain, since the log-spectra can be regarded as scaling factors. The probabilistic relation between frequency coefficients and log-spectra allows these to be treated as two random variables, both to be estimated from the noisy signals. Expectation-maximization (EM) was used to train the GSMM and Bayesian inference was used to compute the posterior signal distribution. Because exact inference of this full probabilistic model is computationally intractable, we developed two approaches to enhance the efficiency: the Laplace method and a variational approximation. The proposed methods were applied to enhance speech corrupted by Gaussian noise and speech-shaped noise (SSN). For both approximations, signals reconstructed from the estimated frequency coefficients provided higher signal-to-noise ratio (SNR) and those reconstructed from the estimated log-spectra produced lower word recognition error rate because the log-spectra fit the inputs to the recognizer better. Our algorithms effectively reduced the SSN, which algorithms based on spectral analysis were not able to suppress.

This paper presents a new approximate Bayesian estimator for enhancing a noisy speech signal. The speech model is assumed to be a Gaussian mixture model (GMM) in the log-spectral domain. This is in contrast to most current models in frequency domain. Exact signal estimation is a computationally intractable problem. We derive three approximations to enhance the efficiency of signal estimation. The Gaussian approximation transforms the log-spectral domain GMM into the frequency domain using minimal Kullback-Leiber (KL)-divergency criterion. The frequency domain Laplace method computes the maximum a posteriori (MAP) estimator for the spectral amplitude. Correspondingly, the log-spectral domain Laplace method computes the MAP estimator for the log-spectral amplitude. Further, the gain and noise spectrum adaptation are implemented using the expectation-maximization (EM) algorithm within the GMM under Gaussian approximation. The proposed algorithms are evaluated by applying them to enhance the speeches corrupted by the speech-shaped noise (SSN). The experimental results demonstrate that the proposed algorithms offer improved signal-to-noise ratio, lower word recognition error rate, and less spectral distortion.

Pharmacologically isolated GABAergic irregular spiking and stuttering interneurons in the mouse visual cortex display highly irregular spike times, with high coefficients of variation approximately 0.9-3, in response to a depolarizing, constant current input. This is in marked contrast to cortical pyramidal cells, which spike quite regularly in response to the same current injection. We applied time-series analysis methods to show that the irregular behavior of the interneurons was not a consequence of low-dimensional, deterministic processes. These methods were also applied to the Hindmarsh and Rose neuronal model to confirm that the methods are adequate for the types of data under investigation. This result has important consequences for the origin of fluctuations observed in the cortex in vivo.

Neural circuits exploit numerous strategies for encoding information. Although the functional significance of individual coding mechanisms has been investigated, ways in which multiple mechanisms interact and integrate are not well understood. The locust olfactory system, in which dense, transiently synchronized spike trains across ensembles of antenna lobe (AL) neurons are transformed into a sparse representation in the mushroom body (MB; a region associated with memory), provides a well-studied preparation for investigating the interaction of multiple coding mechanisms. Recordings made in vivo from the insect MB demonstrated highly specific responses to odors in Kenyon cells (KCs). Typically, only a few KCs from the recorded population of neurons responded reliably when a specific odor was presented. Different odors induced responses in different KCs. Here, we explored with a biologically plausible model the possibility that a form of plasticity may control and tune synaptic weights of inputs to the mushroom body to ensure the specificity of KCs' responses to familiar or meaningful odors. We found that plasticity at the synapses between the AL and the MB efficiently regulated the delicate tuning necessary to selectively filter the intense AL oscillatory output and condense it to a sparse representation in the MB. Activity-dependent plasticity drove the observed specificity, reliability, and expected persistence of odor representations, suggesting a role for plasticity in information processing and making a testable prediction about synaptic plasticity at AL-MB synapses.

Transduction of graded synaptic input into trains of all-or-none action potentials (spikes) is a crucial step in neural coding. Hodgkin identified three classes of neurons with qualitatively different analog-to-digital transduction properties. Despite widespread use of this classification scheme, a generalizable explanation of its biophysical basis has not been described. We recorded from spinal sensory neurons representing each class and reproduced their transduction properties in a minimal model. With phase plane and bifurcation analysis, each class of excitability was shown to derive from distinct spike initiating dynamics. Excitability could be converted between all three classes by varying single parameters; moreover, several parameters, when varied one at a time, had functionally equivalent effects on excitability. From this, we conclude that the spike-initiating dynamics associated with each of Hodgkin's classes represent different outcomes in a nonlinear competition between oppositely directed, kinetically mismatched currents. Class 1 excitability occurs through a saddle node on invariant circle bifurcation when net current at perithreshold potentials is inward (depolarizing) at steady state. Class 2 excitability occurs through a Hopf bifurcation when, despite net current being outward (hyperpolarizing) at steady state, spike initiation occurs because inward current activates faster than outward current. Class 3 excitability occurs through a quasi-separatrix crossing when fast-activating inward current overpowers slow-activating outward current during a stimulus transient, although slow-activating outward current dominates during constant stimulation. Experiments confirmed that different classes of spinal lamina I neurons express the subthreshold currents predicted by our simulations and, further, that those currents are necessary for the excitability in each cell class. Thus, our results demonstrate that all three classes of excitability arise from a continuum in the direction and magnitude of subthreshold currents. Through detailed analysis of the spike-initiating process, we have explained a fundamental link between biophysical properties and qualitative differences in how neurons encode sensory input.

MCell is a Monte Carlo simulator of cell microphysiology, and the scalable variant can be used to study challenging problems of interest to the biological community. MCell can currently model a single synapse out of thousands on a single cell. Petascale technology will enable significant advances in the ability to treat larger structures involving many synapses, with correspondingly more complex behavior. However, there are significant challenges to scaling MCell across two orders of magnitude in performance: increased communication delays and uneven workload concentrations. We discuss software solutions currently under investigation that will accompany us on the path to petascale cell microphysiology.

Gaussian scale mixture models offer a top-down description of signal generation that captures key bottom-up statistical characteristics of filter responses to images. However, the pattern of dependence among the filters for this class of models is prespecified. We propose a novel extension to the gaussian scale mixture model that learns the pattern of dependence from observed inputs and thereby induces a hierarchical representation of these inputs. Specifically, we propose that inputs are generated by gaussian variables (modeling local filter structure), multiplied by a mixer variable that is assigned probabilistically to each input from a set of possible mixers. We demonstrate inference of both components of the generative model, for synthesized data and for different classes of natural images, such as a generic ensemble and faces. For natural images, the mixer variable assignments show invariances resembling those of complex cells in visual cortex; the statistics of the gaussian components of the model are in accord with the outputs of divisive normalization models. We also show how our model helps interrelate a wide range of models of image statistics and cortical processing.

The misjudgement of tilt in images lies at the heart of entertaining visual illusions and rigorous perceptual psychophysics. A wealth of findings has attracted many mechanistic models, but few clear computational principles. We adopt a Bayesian approach to perceptual tilt estimation, showing how a smoothness prior offers a powerful way of addressing much confusing data. In particular, we faithfully model recent results showing that confidence in estimation can be systematically affected by the same aspects of images that affect bias. Confidence is central to Bayesian modeling approaches, and is applicable in many other perceptual domains. Perceptual anomalies and illusions, such as the misjudgements of motion and tilt evident in so many psychophysical experiments, have intrigued researchers for decades.13 A Bayesian view48 has been particularly influential in models of motion processing, treating such anomalies as the normative product of prior information (often statistically codifying Gestalt laws) with likelihood information from the actual scenes presented. Here, we expand the range of statistically normative accounts to tilt estimation, for which there are classes of results (on estimation confidence) that are so far not available for motion. The tilt illusion arises when the perceived tilt of a center target is misjudged (ie bias) in the presence of flankers. Another phenomenon, called Crowding, refers to a loss in the confidence (ie sensitivity) of perceived target tilt in the presence of flankers. Attempts have been made to formalize these phenomena quantitatively. Crowding has been modeled as compulsory feature pooling (ie averaging of orientations), ignoring spatial positions.9, 10 The tilt illusion has been explained by lateral interactions11, 12 in populations of orientationtuned units; and by calibration.13 However, most models of this form cannot explain a number of crucial aspects of the data. First, the geometry of the positional arrangement of the stimuli affects attraction versus repulsion in bias, as emphasized by Kapadia et al14 (figure 1A), and others.15, 16 Second, Solomon et al. recently measured bias and sensitivity simultaneously.11 The rich and surprising range of sensitivities, far from flat as a function of flanker angles (figure 1B), are outside the reach of standard models. Moreover, current explanations do not offer a computational account of tilt perception as the outcome of a normative inference process. Here, we demonstrate that a Bayesian framework for orientation estimation, with a prior favoring smoothness, can naturally explain a range of seemingly puzzling tilt data. We explicitly consider both the geometry of the stimuli, and the issue of confidence in the esti- (A) 6 5 4 3 2 1 0 -1 -2

Summary form only given. Biological structures are extremely complex at the cellular level. The MCell project has been highly successful in simulating the microphysiology of systems of modest size, but many larger problems require too much storage and computation time to be simulated on a single workstation. MCell-K, a new parallel variant of MCell, has been implemented using the KeLP framework and is running on NPACl's Blue Horizon. MCell-K not only produces validated results consistent with the serial version of MCell but does so with unprecedented scalability. We have thus found a level of description and a way to simulate cellular systems that can approach the complexity of nature on its own terms. At the heart of MCell is a 3D random walk that models diffusion using a Monte Carlo method. We discuss two challenging issues that arose in parallelizing the diffusion process - detecting time-step termination efficiently and performing parallel diffusion of particles in a biophysically accurate way. We explore the scalability limits of the present parallel algorithm and discuss ways to improve upon these limits.

In the analysis of natural images, Gaussian scale mixtures (GSM) have been used to account for the statistics of (cid:2)lter responses, and to inspire hi- erarchical cortical representational learning schemes. GSMs pose a crit- ical assignment problem, working out which (cid:2)lter responses were gen- erated by a common multiplicative factor. We present a new approach to solving this assignment problem through a probabilistic extension to the basic GSM, and show how to perform inference in the model using Gibbs sampling. We demonstrate the ef(cid:2)cacy of the approach on both synthetic and image data. Understanding the statistical structure of natural images is an important goal for visual neuroscience. Neural representations in early cortical areas decompose images (and likely other sensory inputs) in a way that is sensitive to sophisticated aspects of their probabilistic structure. This structure also plays a key role in methods for image processing and coding. A striking aspect of natural images that has re(cid:3)ections in both top-down and bottom-up modeling is coordination across nearby locations, scales, and orientations. From a top- down perspective, this structure has been modeled using what is known as a Gaussian Scale Mixture model (GSM).1(cid:150)3 GSMs involve a multi-dimensional Gaussian (each di- mension of which captures local structure as in a linear (cid:2)lter), multiplied by a spatialized collection of common hidden scale variables or mixer variables(cid:3) (which capture the coordi- nation). GSMs have wide implications in theories of cortical receptive (cid:2)eld development, eg the comprehensive bubbles framework of Hyv¤arinen.4 The mixer variables provide the top-down account of two bottom-up characteristics of natural image statistics, namely the ‘bowtie’ statistical dependency,5,6 and the fact that the marginal distributions of receptive (cid:2)eld-like (cid:2)lters have high kurtosis.7,8 In hindsight, these ideas also bear a close relation- ship with Ruderman and Bialek’s multiplicative bottom-up image analysis framework9 and statistical models for divisive gain control.6 Coordinated structure has also been addressed in other image work,10(cid:150)14 and in other domains such as speech15 and (cid:2)nance.16 Many approaches to the unsupervised speci(cid:2)cation of representations in early cortical areas rely on the coordinated structure.17(cid:150)21 The idea is to learn linear (cid:2)lters (eg modeling simple cells as in22,23), and then, based on the coordination, to (cid:2)nd combinations of these (perhaps non-linearly transformed) as a way of (cid:2)nding higher order (cid:2)lters (eg complex cells). One critical facet whose speci(cid:2)cation from data is not obvious is the neighborhood arrangement, ie which linear (cid:2)lters share which mixer variables. (cid:3)Mixer variables are also called mutlipliers, but are unrelated to the scales of a wavelet. Here, we suggest a method for (cid:2)nding the neighborhood based on Bayesian inference of the GSM random variables. In section 1, we consider estimating these components based on information from different-sized neighborhoods and show the modes of failure when inference is too local or too global. Based on these observations, in section 2 we propose an extension to the GSM generative model, in which the mixer variables can overlap prob- abilistically. We solve the neighborhood assignment problem using Gibbs sampling, and demonstrate the technique on synthetic data. In section 3, we apply the technique to image data. 1 GSM inference of Gaussian and mixer variables In a simple, n-dimensional, version of a GSM, (cid:2)lter responses l are synthesized y by mul- tiplying an n-dimensional Gaussian with values g = fg1 : : : gng, by a common mixer variable v. (1) We assume g are uncorrelated ((cid:27)2 along diagonal of the covariance matrix). For the ana- lytical calculations, we assume that v has a Rayleigh distribution:

The synchrony of neurons in extrastriate visual cortex is modulated by selective attention even when there are only small changes in firing rate (Fries, Reynolds, Rorie, & Desimone, 2001). We used Hodgkin-Huxley type models of cortical neurons to investigate the mechanism by which the degree of synchrony can be modulated independently of changes in firing rates. The synchrony of local networks of model cortical interneurons interacting through GABA(A) synapses was modulated on a fast timescale by selectively activating a fraction of the interneurons. The activated interneurons became rapidly synchronized and suppressed the activity of the other neurons in the network but only if the network was in a restricted range of balanced synaptic background activity. During stronger background activity, the network did not synchronize, and for weaker background activity, the network synchronized but did not return to an asynchronous state after synchronizing. The inhibitory output of the network blocked the activity of pyramidal neurons during asynchronous network activity, and during synchronous network activity, it enhanced the impact of the stimulus-related activity of pyramidal cells on receiving cortical areas (Salinas & Sejnowski, 2001). Synchrony by competition provides a mechanism for controlling synchrony with minor alterations in rate, which could be useful for information processing. Because traditional methods such as cross-correlation and the spike field coherence require several hundred milliseconds of recordings and cannot measure rapid changes in the degree of synchrony, we introduced a new method to detect rapid changes in the degree of coincidence and precision of spike timing.

Missing data are common in real-world data sets and are a problem for many estimation techniques. We have developed a variational Bayesian method to perform independent component analysis (ICA) on high-dimensional data containing missing entries. Missing data are handled naturally in the Bayesian framework by integrating the generative density model. Modeling the distributions of the independent sources with mixture of gaussians allows sources to be estimated with different kurtosis and skewness. Unlike the maximum likelihood approach, the variational Bayesian method automatically determines the dimensionality of the data and yields an accurate density model for the observed data without overfitting problems. The technique is also extended to the clusters of ICA and supervised classification framework.

Neurons in the early stages of processing in the primate visual system efficiently encode natural scenes. In previous studies of the chromatic properties of natural images, the inputs were sampled on a regular array, with complete color information at every location. However, in the retina cone photoreceptors with different spectral sensitivities are arranged in a mosaic. We used an unsupervised neural network model to analyze the statistical structure of retinal cone mosaic responses to calibrated color natural images. The second-order statistical dependencies derived from the covariance matrix of the sensory signals were removed in the first stage of processing. These decorrelating filters were similar to type I receptive fields in parvo- or konio-cellular LGN in both spatial and chromatic characteristics. In the subsequent stage, the decorrelated signals were linearly transformed to make the output as statistically independent as possible, using independent component analysis. The independent component filters showed luminance selectivity with simple-cell-like receptive fields, or had strong color selectivity with large, often double-opponent, receptive fields, both of which were found in the primary visual cortex (V1). These results show that the "form" and "color" channels of the early visual system can be derived from the statistics of sensory signals.

Algorithms for data-driven learning of domain-specific overcomplete dictionaries are developed to obtain maximum likelihood and maximum a posteriori dictionary estimates based on the use of Bayesian models with concave/Schur-concave (CSC) negative log priors. Such priors are appropriate for obtaining sparse representations of environmental signals within an appropriately chosen (environmentally matched) dictionary. The elements of the dictionary can be interpreted as concepts, features, or words capable of succinct expression of events encountered in the environment (the source of the measured signals). This is a generalization of vector quantization in that one is interested in a description involving a few dictionary entries (the proverbial "25 words or less"), but not necessarily as succinct as one entry. To learn an environmentally adapted dictionary capable of concise expression of signals generated by the environment, we develop algorithms that iterate between a representative set of sparse representations found by variants of FOCUSS and an update of the dictionary using these sparse representations. Experiments were performed using synthetic data and natural images. For complete dictionaries, we demonstrate that our algorithms have improved performance over other independent component analysis (ICA) methods, measured in terms of signal-to-noise ratios of separated sources. In the overcomplete case, we show that the true underlying dictionary and sparse sources can be accurately recovered. In tests with natural images, learned overcomplete dictionaries are shown to have higher coding efficiency than complete dictionaries; that is, images encoded with an overcomplete dictionary have both higher compression (fewer bits per pixel) and higher accuracy (lower mean square error).

We present ongoing work on a project for automatic recognition of spon- taneous facial actions. Spontaneous facial expressions differ substan- tially from posed expressions, similar to how continuous, spontaneous speech differs from isolated words produced on command. Previous methods for automatic facial expression recognition assumed images were collected in controlled environments in which the subjects delib- erately faced the camera. Since people often nod or turn their heads, automatic recognition of spontaneous facial behavior requires methods for handling out-of-image-plane head rotations. Here we explore an ap- proach based on 3-D warping of images into canonical views. We eval- uated the performance of the approach as a front-end for a spontaneous expression recognition system using support vector machines and hidden Markov models. This system employed general purpose learning mech- anisms that can be applied to recognition of any facial movement. The system was tested for recognition of a set of facial actions defined by the Facial Action Coding System (FACS). We showed that 3D tracking and warping followed by machine learning techniques directly applied to the warped images, is a viable and promising technology for automatic facial expression recognition. One exciting aspect of the approach pre- sented here is that information about movement dynamics emerged out of filters which were derived from the statistics of images.

Missing data is common in real-world datasets and is a problem for many estimation techniques. We have developed a variational Bayesian method to perform Independent Component Analysis (ICA) on high-dimensional data containing missing entries. Missing data are handled naturally in the Bayesian framework by integrating the generative density model. Mod- eling the distributions of the independent sources with mixture of Gaus- sians allows sources to be estimated with different kurtosis and skewness. The variational Bayesian method automatically determines the dimen- sionality of the data and yields an accurate density model for the ob- served data without overfitting problems. This allows direct probability estimation of missing values in the high dimensional space and avoids dimension reduction preprocessing which is not feasible with missing data.

Neurons are sensitive to correlations among synaptic inputs. However, analytical models that explicitly include correlations are hard to solve analytically, so their influence on a neuron's response has been difficult to ascertain. To gain some intuition on this problem, we studied the firing times of two simple integrate-and-fire model neurons driven by a correlated binary variable that represents the total input current. Analytic expressions were obtained for the average firing rate and coefficient of variation (a measure of spike-train variability) as functions of the mean, variance, and correlation time of the stochastic input. The results of computer simulations were in excellent agreement with these expressions. In these models, an increase in correlation time in general produces an increase in both the average firing rate and the variability of the output spike trains. However, the magnitude of the changes depends differentially on the relative values of the input mean and variance: the increase in firing rate is higher when the variance is large relative to the mean, whereas the increase in variability is higher when the variance is relatively small. In addition, the firing rate always tends to a finite limit value as the correlation time increases toward infinity, whereas the coefficient of variation typically diverges. These results suggest that temporal correlations may play a major role in determining the variability as well as the intensity of neuronal spike trains.

When periodic current is injected into an integrate-and-fire model neuron, the voltage as a function of time converges from different initial conditions to an attractor that produces reproducible sequences of spikes. The attractor reliability is a measure of the stability of spike trains against intrinsic noise and is quantified here as the inverse of the number of distinct spike trains obtained in response to repeated presentations of the same stimulus. High reliability characterizes neurons that can support a spike-time code, unlike neurons with discharges forming a renewal process (such as a Poisson process). These two classes of responses cannot be distinguished using measures based on the spike-time histogram, but they can be identified by the attractor dynamics of spike trains, as shown here using a new method for calculating the attractor reliability. We applied these methods to spike trains obtained from current injection into cortical neurons recorded in vitro. These spike trains did not form a renewal process and had a higher reliability compared to renewal-like processes with the same spike-time histogram.

Invariant features of temporally varying signals are useful for analysis and classification. Slow feature analysis (SFA) is a new method for learning invariant or slowly varying features from a vectorial input signal. It is based on a nonlinear expansion of the input signal and application of principal component analysis to this expanded signal and its time derivative. It is guaranteed to find the optimal solution within a family of functions directly and can learn to extract a large number of decorrelated features, which are ordered by their degree of invariance. SFA can be applied hierarchically to process high-dimensional input signals and extract complex features. SFA is applied first to complex cell tuning properties based on simple cell output, including disparity and motion. Then more complicated input-output functions are learned by repeated application of SFA. Finally, a hierarchical network of SFA modules is presented as a simple model of the visual system. The same unstructured network can learn translation, size, rotation, contrast, or, to a lesser degree, illumination invariance for one-dimensional objects, depending on only the training stimulus. Surprisingly, only a few training objects suffice to achieve good generalization to new objects. The generated representation is suitable for object recognition. Performance degrades if the network is trained to learn multiple invariances simultaneously.

A number of current face recognition algorithms use face representations found by unsupervised statistical methods. Typically these methods find a set of basis images and represent faces as a linear combination of those images. Principal component analysis (PCA) is a popular example of such methods. The basis images found by PCA depend only on pairwise relationships between pixels in the image database. In a task such as face recognition, in which important information may be contained in the high-order relationships among pixels, it seems reasonable to expect that better basis images may be found by methods sensitive to these high-order statistics. Independent component analysis (ICA), a generalization of PCA, is one such method. We used a version of ICA derived from the principle of optimal information transfer through sigmoidal neurons. ICA was performed on face images in the FERET database under two different architectures, one which treated the images as random variables and the pixels as outcomes, and a second which treated the pixels as random variables and the images as outcomes. The first architecture found spatially local basis images for the faces. The second architecture produced a factorial face code. Both ICA representations were superior to representations based on PCA for recognizing faces across days and changes in expression. A classifier that combined the two ICA representations gave the best performance.

When a flash is aligned with a moving object, subjects perceive the flash to lag behind the moving object. Two different models have been proposed to explain this "flash-lag" effect. In the motion extrapolation model, the visual system extrapolates the location of the moving object to counteract neural propagation delays, whereas in the latency difference model, it is hypothesized that moving objects are processed and perceived more quickly than flashed objects. However, recent psychophysical experiments suggest that neither of these interpretations is feasible (Eagleman & Sejnowski, 2000a, 2000b, 2000c), hypothesizing instead that the visual system uses data from the future of an event before committing to an interpretation. We formalize this idea in terms of the statistical framework of optimal smoothing and show that a model based on smoothing accounts for the shape of psychometric curves from a flash-lag experiment involving random reversals of motion direction. The smoothing model demonstrates how the visual system may enhance perceptual accuracy by relying not only on data from the past but also on data collected from the immediate future of an event.

A spike-timing-dependent Hebbian mechanism governs the plasticity of recurrent excitatory synapses in the neocortex: synapses that are activated a few milliseconds before a postsynaptic spike are potentiated, while those that are activated a few milliseconds after are depressed. We show that such a mechanism can implement a form of temporal difference learning for prediction of input sequences. Using a biophysical model of a cortical neuron, we show that a temporal difference rule used in conjunction with dendritic backpropagating action potentials reproduces the temporally asymmetric window of Hebbian plasticity observed physio-logically. Furthermore, the size and shape of the window vary with the distance of the synapse from the soma. Using a simple example, we show how a spike-timing-based temporal difference learning rule can allow a network of neocortical neurons to predict an input a few milliseconds before the input's expected arrival.

The analysis of electroencephalographic (EEG) and magnetoencephalographic (MEG) recordings is important both for basic brain research and for medical diagnosis and treatment. Independent component analysis (ICA) is an effective method for removing artifacts and separating sources of the brain signals from these recordings. A similar approach is proving useful for analyzing functional magnetic resonance brain imaging (fMRI) data. In this paper, we outline the assumptions underlying ICA and demonstrate its application to a variety of electrical and hemodynamic recordings from the human brain.

We examine eight different techniques for developing visual rep(cid:173) resentations in machine vision tasks. In particular we compare different versions of principal component and independent com(cid:173) ponent analysis in combination with stepwise regression methods for variable selection. We found that local methods, based on the statistics of image patches, consistently outperformed global meth(cid:173) ods based on the statistics of entire images. This result is consistent with previous work on emotion and facial expression recognition. In addition, the use of a stepwise regression technique for selecting variables and regions of interest substantially boosted performance.

The human visual system encodes the chromatic signals conveyed by the three types of retinal cone photoreceptors in an opponent fashion. This color opponency has been shown to constitute an efficient encoding by spectral decorrelation of the receptor signals. We analyze the spatial and chromatic structure of natural scenes by decomposing the spectral images into a set of linear basis functions such that they constitute a representation with minimal redun(cid:173) dancy. Independent component analysis finds the basis functions that transforms the spatiochromatic data such that the outputs (activations) are statistically as independent as possible, i.e. least redundant. The resulting basis functions show strong opponency along an achromatic direction (luminance edges), along a blue(cid:173) yellow direction, and along a red-blue direction. Furthermore, the resulting activations have very sparse distributions, suggesting that the use of color opponency in the human visual system achieves a highly efficient representation of colors. Our findings suggest that color opponency is a result of the properties of natural spectra and not solely a consequence of the overlapping cone spectral sensitiv(cid:173) ities. 1 Statistical structure of natural scenes Efficient encoding of visual sensory information is an important task for informa(cid:173) tion processing systems and its study may provide insights into coding principles of biological visual systems. An important goal of sensory information processing Electronic version available at www. cnl. salk . edu/ ""tewon. is to transform the input signals such that the redundancy between the inputs is reduced. In natural scenes, the image intensity is highly predictable from neighbor(cid:173) ing measurements and an efficient representation preserves the information while the neuronal output is minimized. Recently, several methods have been proposed for finding efficient codes for achromatic images of natural scenes [1, 2, 3, 4]. While luminance dominates the structure of the visual world, color vision provides impor(cid:173) tant additional information about our environment. Therefore, we are interested in efficient, i.e. redundancy reducing representations for the chromatic structure of natural scenes. 2 Learning efficient representation for chromatic image Our goal was to find efficient representations of the chromatic sensory information such that its spatial and chromatic redundancy is reduced significantly. The method we used for finding statistically efficient representations is independent component analysis (ICA). ICA is a way of finding a linear non-orthogonal co-ordinate system in multivariate data that minimizes mutual information among the axial projections of the data. The directions of the axes of this co-ordinate system (basis functions) are determined by both second and higher-order statistics of the original data, com(cid:173) pared to Principal Component Analysis (PCA) which is used solely in second order statistics and has orthogonal basis functions. The goal of ICA is to perform a linear transform which makes the resulting source outputs as statistically indepen(cid:173) dent from each other as possible [5]. ICA assumes an unknown source vector s with mutually independent components Si. A small patch of the observed image is stretched into a vector x that can be represented as a linear combination of sources components Si such that

In an overcomplete basis, the number of basis vectors is greater than the dimensionality of the input, and the representation of an input is not a unique combination of basis vectors. Overcomplete representations have been advocated because they have greater robustness in the presence of noise, can be sparser, and can have greater flexibility in matching structure in the data. Overcomplete codes have also been proposed as a model of some of the response properties of neurons in primary visual cortex. Previous work has focused on finding the best representation of a signal using a fixed overcomplete basis (or dictionary). We present an algorithm for learning an overcomplete basis by viewing it as probabilistic model of the observed data. We show that overcomplete bases can yield a better approximation of the underlying statistical distribution of the data and can thus lead to greater coding efficiency. This can be viewed as a generalization of the technique of independent component analysis and provides a method for Bayesian reconstruction of signals in the presence of noise and for blind source separation when there are more sources than mixtures.

An unsupervised classification algorithm is derived by modeling observed data as a mixture of several mutually exclusive classes that are each described by linear combinations of independent, non-Gaussian densities. The algorithm estimates the density of each class and is able to model class distributions with non-Gaussian structure. The new algorithm can improve classification accuracy compared with standard Gaussian mixture models. When applied to blind source separation in nonstationary environments, the method can switch automatically between classes, which correspond to contexts with different mixing properties. The algorithm can learn efficient codes for images containing both natural scenes and text. This method shows promise for modeling non-Gaussian structure in high-dimensional data and has many potential applications.

A recent study of cat visual cortex reported abrupt changes in the positions of the receptive fields of adjacent neurons whose preferred orientations strongly differed (Das & Gilbert, 1997). Using a simple cortical model, we show that this covariation of discontinuities in maps of orientation preference and local distortions in maps of visual space reflects collective effects of the lateral cortical feedback.

An extension of the infomax algorithm of Bell and Sejnowski (1995) is presented that is able blindly to separate mixed signals with sub- and supergaussian source distributions. This was achieved by using a simple type of learning rule first derived by Girolami (1997) by choosing negentropy as a projection pursuit index. Parameterized probability distributions that have sub- and supergaussian regimes were used to derive a general learning rule that preserves the simple architecture proposed by Bell and Sejnowski (1995), is optimized using the natural gradient by Amari (1998), and uses the stability analysis of Cardoso and Laheld (1996) to switch between sub- and supergaussian regimes. We demonstrate that the extended infomax algorithm is able to separate 20 sources with a variety of source distributions easily. Applied to high-dimensional data from electroencephalographic recordings, it is effective at separating artifacts such as eye blinks and line noise from weaker electrical signals that arise from sources in the brain.

When the nervous system is presented with multiple simultaneous inputs of some variable, such as wavelength or disparity, they can be combined to give rise to qualitatively new percepts that cannot be produced by any single input value. For example, there is no single wavelength that appears white. Many models of decoding neural population codes have problems handling multiple inputs, either attempting to extract a single value of the input parameter or, in some cases, registering the presence of multiple inputs without synthesizing them into something new. These examples raise a more general issue regarding the interpretation of population codes. We propose that population decoding involves not the extraction of specific values of the physical inputs, but rather a transformation from the input space to some abstract representational space that is not simply related to physical parameters. As a specific example, a four-layer network is presented that implements a transformation from wavelength to a high-level hue-saturation color space.

Sensory and motor variables are typically represented by a population of broadly tuned neurons. A coarser representation with broader tuning can often improve coding accuracy, but sometimes the accuracy may also improve with sharper tuning. The theoretical analysis here shows that the relationship between tuning width and accuracy depends crucially on the dimension of the encoded variable. A general rule is derived for how the Fisher information scales with the tuning width, regardless of the exact shape of the tuning function, the probability distribution of spikes, and allowing some correlated noise between neurons. These results demonstrate a universal dimensionality effect in neural population coding.

The Facial Action Coding System (FACS) [23] is an objective method for quantifying facial movement in terms of component actions. This system is widely used in behavioral investigations of emotion, cognitive processes, and social interaction. The coding is presently performed by highly trained human experts. This paper explores and compares techniques for automatically recognizing facial actions in sequences of images. These techniques include analysis of facial motion through estimation of optical flow; holistic spatial analysis, such as principal component analysis, independent component analysis, local feature analysis, and linear discriminant analysis; and methods based on the outputs of local filters, such as Gabor wavelet representations and local principal components. Performance of these systems is compared to naive and expert human subjects. Best performances were obtained using the Gabor wavelet representation and the independent component representation, both of which achieved 96 percent accuracy for classifying 12 facial actions of the upper and lower face. The results provide converging evidence for the importance of using local filters, high spatial frequencies, and statistical independence for classifying facial actions.

Empirical results were obtained for the blind source separation of more sources than mixtures using a previously proposed framework for learning overcomplete representations. This technique assumes a linear mixing model with additive noise and involves two steps: (1) learning an overcomplete representation for the observed data and (2) inferring sources given a sparse prior on the coefficients. We demonstrate that three speech signals can be separated with good fidelity given only two mixtures of the three signals. Similar results were obtained with mixtures of two speech signals and one music signal.

We present methods to separate blindly mixed signals recorded in a room. The learning algorithm is based on the information maximization in a single layer neural network. We focus on the implementation of the learning algorithm and on issues that arise when separating speakers in room recordings. We used an infomax approach in a feedforward neural network implemented in the frequency domain using the polynomial filter matrix algebra technique. A fast convergence speed was achieved by using a time-delayed decorrelation method as a preprocessing step. Under minimum-phase mixing conditions this preprocessing step was sufficient for the separation of signals. These methods successfully separated a recorded voice with music in the background (cocktail party problem). Finally, we discuss problems that arise in real world recordings and their potential solutions.

Computational models of neural map formation can be considered on at least three different levels of abstraction: detailed models including neural activity dynamics, weight dynamics that abstract from the neural activity dynamics by an adiabatic approximation, and constrained optimization from which equations governing weight dynamics can be derived. Constrained optimization uses an objective function, from which a weight growth rule can be derived as a gradient flow, and some constraints, from which normalization rules are derived. In this article, we present an example of how an optimization problem can be derived from detailed nonlinear neural dynamics. A systematic investigation reveals how different weight dynamics introduced previously can be derived from two types of objective function terms and two types of constraints. This includes dynamic link matching as a special case of neural map formation. We focus in particular on the role of coordinate transformations to derive different weight dynamics from the same optimization problem. Several examples illustrate how the constrained optimization framework can help in understanding, generating, and comparing different models of neural map formation. The techniques used in this analysis may also be useful in investigating other types of neural dynamics.

Many different algorithms and objective functions for topographic mappings have been proposed. We show that several of these approaches can be seen as particular cases of a more general objective function. Consideration of a very simple mapping problem reveals large differences in the form of the map that each particular case favors. These differences have important consequences for the practical application of topographic mapping methods.

established from the perspective of psychological experiments, the neural mechanisms that underlie this pre-

Blind separation and blind deconvolution are related problems in unsupervised learning. In this contribution, static non-linearities are used in combination with an information-theoretic objective function, making the approach more rigorous than previous ones. We derive a new algorithm and with it perform nearly perfect separation of up to 10 digitally mixed human speakers, better performance than any previous algorithms for blind separation. When used for deconvolution, the technique automatically cancels echoes and reverberations and reverses the effects of low-pass filtering.

We derive a new self-organizing learning algorithm that maximizes the information transferred in a network of nonlinear units. The algorithm does not assume any knowledge of the input distributions, and is defined here for the zero-noise limit. Under these conditions, information maximization has extra properties not found in the linear case (Linsker 1989). The nonlinearities in the transfer function are able to pick up higher-order moments of the input distributions and perform something akin to true redundancy reduction between units in the output representation. This enables the network to separate statistically independent components in the inputs: a higher-order generalization of principal components analysis. We apply the network to the source separation (or cocktail party) problem, successfully separating unknown mixtures of up to 10 speakers. We also show that a variant on the network architecture is able to perform blind deconvolution (cancellation of unknown echoes and reverberation in a speech signal). Finally, we derive dependencies of information transfer on time delays. We suggest that information maximization provides a unifying framework for problems in "blind" signal processing.

ermine how the reliability of sex discrimination is related to resolution. A normalized pixel-based representation was used for the faces because it explicitly retained texture and shape information while also maintaining geometric relationships. We found that the linear perceptron model can classify sex from facial images with 81% accuracy, compared to 92% accuracy with compression coding on the same data set [6]. The advantage of using a simple linear perceptron with normalized pixel-based inputs is that it allows us to see explicitly those regions of the face that make the largest and most reliable contributions to the classification of sex. A database of 90 faces (44 males, 46 females) was used (O'Toole, Millward, & Anderson [9]). No facial hair, jewelry, or makeup was on any of the faces. Each face was rotated until the eyes were level, and then scaled and cropped so that each image showed a similar facial area. From the original set of faces, we created 5 separate databases at 5

A new learning algorithm is derived which performs online stochas(cid:173) tic gradient ascent in the mutual information between outputs and inputs of a network. In the absence of a priori knowledge about the 'signal' and 'noise' components of the input, propagation of information depends on calibrating network non-linearities to the detailed higher-order moments of the input density functions. By incidentally minimising mutual information between outputs, as well as maximising their individual entropies, the network 'fac(cid:173) torises' the input into independent components. As an example application, we have achieved near-perfect separation of ten digi(cid:173) tally mixed speech signals. Our simulations lead us to believe that our network performs better at blind separation than the Herault(cid:173) J utten network, reflecting the fact that it is derived rigorously from the mutual information objective. 468 Anthony J. Bell, Terrence J. Sejnowski

Songbirds learn to imitate a tutor song through auditory and motor learn(cid:173) ing. We have developed a theoretical framework for song learning that accounts for response properties of neurons that have been observed in many of the nuclei that are involved in song learning. Specifically, we suggest that the anteriorforebrain pathway, which is not needed for song production in the adult but is essential for song acquisition, provides synaptic perturbations and adaptive evaluations for syllable vocalization learning. A computer model based on reinforcement learning was con(cid:173) structed that could replicate a real zebra finch song with 90% accuracy based on a spectrographic measure. The second generation of the bird(cid:173) song model replicated the tutor song with 96% accuracy.

The auditory system of the barn owl contains several spatial maps. In young barn owls raised with optical prisms over their eyes, these auditory maps are shifted to stay in register with the visual map, suggesting that the visual input imposes a frame of reference on the auditory maps. However, the optic tectum, the first site of convergence of visual with auditory information, is not the site of plasticity for the shift of the auditory maps; the plasticity occurs instead in the inferior colliculus, which contains an auditory map and projects into the optic tectum. We explored a model of the owl remapping in which a global reinforcement signal whose delivery is controlled by visual foveation. A hebb learning rule gated by rein(cid:173) forcement learned to appropriately adjust auditory maps. In addi(cid:173) tion, reinforcement learning preferentially adjusted the weights in the inferior colliculus, as in the owl brain, even though the weights were allowed to change throughout the auditory system. This ob(cid:173) servation raises the possibility that the site of learning does not have to be genetically specified, but could be determined by how the learning procedure interacts with the network architecture. 126 Alexandre Pouget, Cedric Deffayet, Te"ence J. Sejnowski c:::======:::::» •

The parietal cortex is thought to represent the egocentric posi(cid:173) tions of objects in particular coordinate systems. We propose an alternative approach to spatial perception of objects in the pari(cid:173) etal cortex from the perspective of sensorimotor transformations. The responses of single parietal neurons can be modeled as a gaus(cid:173) sian function of retinal position multiplied by a sigmoid function of eye position, which form a set of basis functions. We show here how these basis functions can be used to generate receptive fields in either retinotopic or head-centered coordinates by simple linear transformations. This raises the possibility that the parietal cortex does not attempt to compute the positions of objects in a partic(cid:173) ular frame of reference but instead computes a general purpose representation of the retinal location and eye position from which any transformation can be synthesized by direct projection. This representation predicts that hemineglect, a neurological syndrome produced by parietal lesions, should not be confined to egocentric coordinates, but should be observed in multiple frames of reference in single patients, a prediction supported by several experiments. 158 Alexandre Pouget, Terrence J. Sejnowski

Differentiation between the nodes of a competitive learning net(cid:173) work is conventionally achieved through competition on the ba(cid:173) sis of neural activity. Simple inhibitory mechanisms are limited to sparse representations, while decorrelation and factorization schemes that support distributed representations are computation(cid:173) ally unattractive. By letting neural plasticity mediate the compet(cid:173) itive interaction instead, we obtain diffuse, nonadaptive alterna(cid:173) tives for fully distributed representations. We use this technique to Simplify and improve our binary information gain optimiza(cid:173) tion algorithm for feature extraction (Schraudolph and Sejnowski, 1993); the same approach could be used to improve other learning algorithms.

Diagnosis of human disease or machine fault is a missing data problem since many variables are initially unknown. Additional information needs to be obtained. The j oint probability distribution of the data can be used to solve this problem. We model this with mixture models whose parameters are estimated by the EM algorithm. This gives the benefit that missing data in the database itself can also be handled correctly. The request for new information to refine the diagnosis is performed using the maximum utility principle. Since the system is based on learning it is domain independent and less labor intensive than expert systems or probabilistic networks. An example using a heart disease database is presented.

Many cells in the dorsal part of the medial superior temporal (MST) area of visual cortex respond selectively to spiral flow patterns-specific combinations of expansion/ contraction and ro(cid:173) tation motions. Previous investigators have suggested that these cells may represent self-motion. Spiral patterns can also be gener(cid:173) ated by the relative motion of the observer and a particular object. An MST cell may then account for some portion of the complex flow field, and the set of active cells could encode the entire flow; in this manner, MST effectively segments moving objects. Such a grouping operation is essential in interpreting scenes containing several independent moving objects and observer motion. We de(cid:173) scribe a model based on the hypothesis that the selective tuning of MST cells reflects the grouping of object components undergo(cid:173) ing coherent motion. Inputs to the model were generated from sequences of ray-traced images that simulated realistic motion sit(cid:173) uations, combining observer motion, eye movements, and indepen(cid:173) dent object motion. The input representation was modeled after response properties of neurons in area MT, which provides the pri(cid:173) mary input to area MST. After applying an unsupervised learning algorithm, the units became tuned to patterns signaling coherent motion. The results match many of the known properties of MST cells and are consistent with recent studies indicating that these cells process 3-D object motion information. 166 Richard S. Zemel, Terrence J. Sejnowski

International audience

Hebbian synapses lie at the heart of most associative matrix memories (Kohonen 1987; Hinton and Anderson 1981) and are also biologically plausible (Brown et al. 1990; Baudry and Davis 1991). Their analytical and computational tractability make these memories the best understood form of distributed information storage. A variety of Hebbian algorithms for estimating the covariance between input and output patterns has been proposed. This note points out that one class of these involves stochastic estimation of the covariance, shows that the signal-to-noise ratios of the rules are governed by the variances of their estimates, and considers some parallels in reinforcement learning.

When cerebellar Purkinje cells are depolarized with a constant current pulse injected at the soma, complex spike discharge patterns are observed (Llinas and Sugimori 1980b). A computer model has been constructed to analyze how the Purkinje cell ionic conductance identified to date interact to produce the observed firing behavior. The kinetics of voltage-dependent conductance used in the model were significantly simpler than Hodgkin-Huxley kinetics, which have many parameters that must be experimentally determined. Our simplified scheme was able to reproduce the complex nonlinear responses found in real Purkinje cells. A similar approach could be used to study the wide variety of neurons found in different brain regions.

Interneurons in leech ganglia receive multiple sensory inputs and make synaptic contacts with many motor neurons. These hidden units coordinate several different behaviors. The authors used physiological and anatomical constraints to construct a model of the local bending reflex. Dynamical networks were trained on experimentally derived input-output patterns using recurrent back propagation. Units in the model were modified to include electrical synapses and multiple synaptic time constants. The properties of the hidden units that emerged in the simulations matched those in the leech. The model and data support distributed rather than localist representations in the local bending reflex. The results also explain counterintuitive aspects of the local bending circuitry

September 01 1990 Faster Learning for Dynamic Recurrent Backpropagation Yan Fang, Yan Fang The Salk Institute, Computational Neurobiology Laboratory, 10010 N. Torrey Pines Road, La Jolla, CA 92037 USA Search for other works by this author on: This Site Google Scholar Terrence J. Sejnowski Terrence J. Sejnowski The Salk Institute, Computational Neurobiology Laboratory, 10010 N. Torrey Pines Road, La Jolla, CA 92037 USA Search for other works by this author on: This Site Google Scholar Author and Article Information Yan Fang The Salk Institute, Computational Neurobiology Laboratory, 10010 N. Torrey Pines Road, La Jolla, CA 92037 USA Terrence J. Sejnowski The Salk Institute, Computational Neurobiology Laboratory, 10010 N. Torrey Pines Road, La Jolla, CA 92037 USA Received: January 08 1990 Accepted: May 23 1990 Online Issn: 1530-888X Print Issn: 0899-7667 © 1990 Massachusetts Institute of Technology1990 Neural Computation (1990) 2 (3): 270–273. https://doi.org/10.1162/neco.1990.2.3.270 Article history Received: January 08 1990 Accepted: May 23 1990 Cite Icon Cite Permissions Share Icon Share Twitter LinkedIn Views Icon Views Article contents Figures & tables Video Audio Supplementary Data Peer Review Search Site Citation Yan Fang, Terrence J. Sejnowski; Faster Learning for Dynamic Recurrent Backpropagation. Neural Comput 1990; 2 (3): 270–273. doi: https://doi.org/10.1162/neco.1990.2.3.270 Download citation file: Ris (Zotero) Reference Manager EasyBib Bookends Mendeley Papers EndNote RefWorks BibTex toolbar search Search Dropdown Menu nav search search input Search input auto suggest search filter All ContentAll JournalsNeural Computation Search Advanced Search This content is only available as a PDF. © 1990 Massachusetts Institute of Technology1990 Article PDF first page preview Close Modal You do not currently have access to this content.

Interneurons in leech ganglia receive multiple sensory inputs and make synaptic contacts with many motor neurons. These “hidden” units coordinate several different behaviors. We used physiological and anatomical constraints to construct a model of the local bending reflex. Dynamic networks were trained on experimentally derived input-output patterns using recurrent backpropagation. Units in the model were modified to include electrical synapses and multiple synaptic time constants. The properties of the hidden units that emerged in the simulations matched those in the leech. The model and data support distributed rather than localist representations in the local bending reflex. These results also explain counterintuitive aspects of the local bending circuitry.

It is demonstrated that multiple sources of speech information can be integrated at a subsymbolic level to improve vowel recognition. Feedforward and recurrent neural networks are trained to estimate the acoustic characteristics of a vocal tract from images of the speaker's mouth. These estimates are then combined with the noise-degraded acoustic information, effectively increasing the signal-to-noise ratio and improving the recognition of these noise-degraded signals. Alternative symbolic strategies such as direct categorization of the visual signals into vowels are also presented. The performances of these neural networks compare favorably with human performance and with other pattern-matching and estimation techniques.>

No abstract available.