1999
1999 Annual Research Report
Terrence J. Sejnowski
Research Projects
Thomas M. Bartol Jr. and Joel R. Stiles (Cornell University)
MCell: A Monte Carlo simulator of cellular microphysiology
MCell is a software program for 3-D Monte Carlo simulation of ligand diffusion and chemical signaling, focusing on neurotransmitter release and quantal current generation at peripheral and central synapses. MCell has been highly optimized and its generality expanded to allow simulation of multiple ligand and receptor classes, along with complex 3-D arrangements of diffusion boundaries representing multiple cell or organelle membranes. Diffusion boundaries may now be constructed from 3-D triangle meshes such as those obtained from serial reconstruction. Simulations are designed using a Model Description Language to define ligands and other molecular constituents (e.g. receptors, enzymes, uptake sites), the arrangement of boundaries, the timing of ligand release, and additional parameters. Thus, many processes in addition to synaptic transmission can now be modeled. MCell is currently in use by 20 laboratories around the world under a beta-test agreement. Some of the projects using MCell described elsewhere in this report include the triggering of transmitter release from presynaptic terminals and the induction of long-term synaptic plasticity in the postsynaptic spine of excitatory synapses. For more details see: http://www.mcell.cnl.salk.edu
Maxim Bazhenov, Igor Timofeev (Laval University, Canada) and Mircea Steriade (Laval University, Canada)
Spontaneous activity in the isolated cortical slab in vivo
Igor Timofeev and Mircea Steriade at Laval University in Quebec have developed a new in vivo preparation to study the mechanisms underlying spontaneous sleep oscillations. Dual and triple simultaneous intracellular recordings were made from neurons in small isolated cortical slabs (10 mm x 6 mm) in anesthetized cats. Spontaneously occurring slow sleep oscillations, present in the adjacent intact cortex, were absent in small slabs. However, the isolated slabs displayed brief active periods separated by long periods of silence, up to 60 s in duration. During these silent periods, 60% of neurons showed non-linear amplification of low amplitude depolarizing activity. Nearly 40% of the cells, twice as many as in intact vortex, were classified as intrinsically bursting. We have explored cortical network models based on Hodgkin-Huxley-like neurons. The summation of simulated spontaneous miniature excitatory postsynaptic potentials was sufficient to activate a persistent sodium current, initiating action potentials in single neurons that then spread through the network. Consistent with this model, enlarging the isolated cortical territory to an isolated gyrus (20 mm x 10 mm) increased the probability of initiating large-scale activity. In these larger territories, both the frequency and regularity of the slow oscillation approached that generated in intact cortex. The frequency of active periods in an analytical model of the cortical network accurately predicted the scaling observed in simulations and from recordings in cortical slabs of increasing size.
Maxim Bazhenov, Mark Stopfer (Caltech), Mikhail Rabinovich (UCSD), Henry D.I. Abarbanel (UCSD) and Gilles Laurent (Caltech)
Cellular and network mechanisms for temporal patterning in the locust antennal lobe
Antennal lobe projection neurons (PNs) display slow temporal patterns of activity during olfactory stimulation. This patterning is a slow alternation of depolarizing and hyperpolarizing activity. The slow temporal structure is stimulus specific and reproducible over the repetitive trials with the same odor. The slow inhibitory receptors between LNs and PNs might provide a mechanism underlying such a temporal patterning. This hypothesis was investigated with computer model including PNs and local neurons (LNs). We found that the activation of the slow inhibitory receptors between LNs and PNs can lead to the relatively slow hyperpolarization some of the PNs and diminishing of the spike activity observed in these cells. Depending on the stimulus-specific temporal patterns of activity in the presynaptic LNs, the hyperpolarizing phases were observed during different temporal epochs of stimulation, which lasted 100 ms to 400 ms. Release from the slow inhibition, after the olfactory stimulation was terminated, led to the growth of activity in many PNs, so their response patterns outlasted the duration of the stimulus. The slow temporal patterns of PNs activity were stimulus specific and showed relatively small changes when the fast inhibitory connections between LNs and from LNs to PNs were blocked. The model predicts that the slow inhibitory receptors can contribute to the temporal coding of olfactory information.
Gary Bedford and Scott Makeig
Independent component analysis of event-related potential (ERP) studies of source monitoring in memory
Human event-related potential (ERP) data recorded in two experiments designed to explore the differences between activations produced by source monitoring and recognition tasks in memory was analyzed using independent component analysis (ICA). These source monitoring experiments sought to determine how differences in task requirements and therefore manner of encoding would affect how and where memories were encoded and subsequently retrieved in the brain. In one experiment, four subject groups each performed three tasks. Each of these groups was clearly distinguished by ICA and separated into several components with three activations each. The scalp maps corresponding to these events, though, while showing some interesting features, failed to show significant differences among the subject groups. It was not possible to say, for example, whether small and potentially interesting differences between subject groups were effects of the small subject pool and subsequent propagation of errors in the averaging of these groups or genuine differences in responses to differing stimuli. Larger subject groups and alterations in experimental design should help resolve these problems. In addition, analyzing single-trial data should overcome the possible errors introduced by averaging.
Glen Brown and Satoru Shiono (Mitsubishi Electric Corporation, Japan)
Analysis of optical recordings from the isolated brain of Tritonia during fictive swimming
We have been analyzing large optical recording data sets from the isolated brain of the seaslug Tritonia. A voltage sensitive dye was used to image many individual neurons during the fictive swimming pattern. Neurons important for swimming are easy to recognize because they burst in phase with the swimming pattern. Previously, analyses of this type required days to weeks of operator time to correct errors made by spike-sorting algorithms. We have applied independent component analysis (ICA) to the problem of isolating spike trains from individual neurons in these recordings. ICA effectively automates the data analysis. Spikes from different neurons are assigned to separate channels even if they change shape during recording, for example during high frequency bursts of spikes. Indeed, ICA returns a continuous estimate of membrane potential so post-synaptic potentials and subthreshold membrane oscillations can also be recovered by this method. ICA also removes artifacts from the data and automatically resolves overlapping action potentials. In the best recordings, more than 100 simultaneously recorded action potential trains were extracted. This nearly doubles the number of neurons known to be involved in the Tritonia swimming neural network.
Stijn Cassener and Thomas Bartol
MCell simulations of calcium entry into presynaptic terminals and the mechanisms underlying paired pulse facilitation and long-term potentiation
The entry of calcium into the presynaptic terminal of chemical synapses serves to trigger the release of neurotransmitter. MCell was used to simulate the gating of single calcium channels. A wide range of experimental observations were used to constrain the model and three predictions were made: 1) Although the entry of calcium through a single channel is sufficient to trigger transmitter release, synchronous release of calcium from overlapping microdomains was needed to reproduce the release probabilities observed experimentally. 2) Paired pulse facilitation occurs when calcium-binding proteins become saturated and the effective concentration of free calcium increase. 3) Long-term potentiation should have two components: a redistribution of probability caused by increased calcium influx, and uniform gain increase owing to an increase in the readily releasable pool of neurotransmitter.
Kwokleung Chan, Te-won Lee and Scott Makeig
Latent sematic indexing (LSI) by independent component analysis (ICA)
Independent component analysis (ICA) was used to explore query-based information retrieval The amount of information instantly available to us has grown tremendously and new methods for automatic information retrieval are needed that can effectively search the data. In this project, we implement ICA was used to implement Latent Sematic Indexing (LSI) and perform a query-based documents retrieval. We compared the ICA performance with Singular Value Decomposition (SVD). For several large datasets, ICA performed well in grouping words together according their semantic meanings, and finding independent topics and concepts from a set of documents. However, as measured by the precision and recall capability test, ICA and SVD perform equally well.
Brian Christie
Synaptic plasticity in hippocampal stratum radiatum interneuron
Hippocampal synaptic plasticity is used as an animal model of the biological processes that underlie learning and memory in the mammalian brain. This phenomenon has been well documented in hippocampal excitatory principal cells that project to other cortical regions, but remain elusive in the inhibitory interneurons that project mainly within the hippocampal formation. The inhibitory neuronal population, while only composing approximately twenty percent of the total hippocampal neuronal pool, appears to play a crucial role in regulating complex interactions of hippocampal excitatory cells, including oscillatory activity, epileptic synchronization, and synaptic plasticity. Although there is considerable morphological diversification of interneurons, in general, each interneuron is capable of influencing the electrical activity of thousands of excitatory cells. Inhibitory interneurons thus have a pivotal position in regulating overall hippocampal excitability. Differential Interference Contrast (DIC) optics was used to directly visualize individual interneurons in the stratum radiatum of the hippocampal formation. Using whole-cell recording techniques, synaptically generated responses are recorded from interneurons prior to and following the application of either LTP-inducing (100 pulses at 200 Hz) or LTD-inducing (900 pulses at 3 Hz) excitatory synaptic stimulation. After recording, each interneuron was filled with biocytin for post-recording histological verification of the morphology and position in the stratum radiatum. Several types of interneurons in the hippocampal stratum radiatum exhibited a form of long-term potentiation (LTP) that requires NMDA-receptor activation but the optimal stimulus was different for each cell type .
Brian Christie, Henriette Van Praag (Salk) and Fred H. Gage (Salk)
Voluntary exercise increases brain neurogenesis, synaptic plasticity, and memory retention
The addition of a running wheel to the cages of C57 mice can markedly enhance the proliferation and density of neurons in adult mice. The objective of this research was to find physiological correlates of these anatomical changes in the dentate gyrus subregion of the hippocampal formation. Mice were randomly placed in cages that either contained a running wheel or did not, and left in these conditions for a period of several weeks. The mice were then tested for learning performance and memory retention using the Morris water-maze. Mice in the running group learned the task in the same manner as those in the control group; however, they retained platform location information significantly longer than the control mice. These same mice were then delivered to a second investigator who was blind as to the conditions the mice were raised in. Field recordings were made from dentate granule cells and paired-pulse facilitation and synaptic plasticity were examined. Surprisingly, mice raised in cages that contained running wheels exhibited significantly more long-term potentiation than those raised under control conditions. Histological analysis of these slices also revealed that they contained significantly more neurons than control slices.
Olivier J.-M. D. Coenen (San Diego Children's Hospital Research Center), Marwan A. Jabri (University of Sydney) and Jerry Huang (University of Sydney)
Sensorimotor integration and control
Recent models of sensorimotor integration and control of the central nervous system are being tested and extended on real-time robotic platforms. The models attempt to represent the interactions between the pre-frontal cortex, the basal ganglia, the ventral tegmental area and the cerebellum. Computational experiments have been performed to study learning and performance in simple tasks, such as tracking and obstacle avoidance. A Khepera microrobot equipped with infra-red proximity sensors and a Nomad robot equipped with sonar sensors and cameras are being used. All computations are performed on computer and commands are communicated to the robot via a serial line. Learning and motor decisions are performed in real-time within a multi-threaded environment. Two sensorimotor integration models have so far been studied. The first implements a simple decision making system comprising models originally derived earlier in the laboratory for the interactions between pre-frontal cortex, basal ganglia and ventral tegmental area. These models were inspired by temporal difference learning and in our current implementation use sensory inputs and previous motor commands to make the next motor commands and reward prediction. The second model augments the first one by including a simple predictor of the robot's environment, modeled abstractly on the cerebellum. The output of the cerebellum model is used as an additional input to networks used in the first model. Our experiments show that the inclusion of the predictions of the environment greatly improves the performance of the robot in tracking external sources by coordinating and smoothing the robot movements. The models will be extended, new tasks will be explored and the performance will be analyzed to understand the crucial elements responsible for successful real-time interactions with the environment. These experiments may improve our understanding of the interactions that occur between different brain areas during complex sensorimotor tasks.
Olivier J.-M. D. Coenen (San Diego Children's Hospital Research Center), Mike Arnold (University of Sydney) and Marwan A. Jabri (University of Sydney)
A hypothesis for parallel fiber coding in the cerebellum
There are more granule cells than any other cell type in the central nervous system. Mossy fiber inputs to the cerebellum terminate in glomeruli where granule cell dendrites and Golgi cell axons converge to make synaptic contacts. Recent experimental results have demonstrated that several neurotransmitters and signaling mechanisms interact in complex ways at the glomeruli. The hypothesis that is explored in this study is that the convergence of inputs on granule cells produces an output that approximates a sparsely-distributed and statistically-independent code. Such a code is beneficial for learning downstream at the Purkinje cells due to the convergence of climbing and parallel fiber activities. The influence of Golgi cells on computation in granule cell is also explored and learning rules for synaptic adaptation are derived using a maximum likelihood method. This work demonstrates that powerful unsupervised computation can take place in the input layer of the cerebellum to facilitate learning at later stages and suggests that synapses between the mossy fibers and the granule cells in the cerebellum uncovers latent structures in the inputs.
David M. Eagleman
Motion integration and postdiction in visual awareness
The flash-lag effect is a robust visual illusion wherein a flash and a moving object that appear in the same location are perceived to be displaced from one another. Two explanations have been suggested: the first proposal is that the visual system is predictive, accounting for neural delays by extrapolating the trajectory of a moving stimulus into the future. The second is that the visual system processes moving objects more quickly than flashed objects. This "latency-difference" hypothesis asserts that by the time the flashed object is processed, the moving object has already moved to a new position. The latter proposal tacitly rests on the assumption that awareness (what the subject reports) is an on-line phenomenon, coming about as soon as a stimulus reaches its "perceptual end-point". We have designed a series of psychophysical experiments to directly test these two frameworks. The results are inconsistent with either proposal. Instead, we propose that visual awareness is postdictive, such that the percept attributed to the time of an event is a function of what happens in the ~80 msec following the event.
Michael Eisele
Learning rules for reinforcement and prediction
The goal of this research is to find synaptic learning rules that agree with known properties of selected brain areas. Computationally simple, but biologically plausible learning rules are simulated and compared with known properties of dopamine neurons in the midbrain. These neurons can learn to predict rewarding stimuli based on eligibility traces, which can be implemented in single neurons. Models of prediction in thalamic and cortical circuits have also been developed based on simple learning rules that are robust to distracting signals. It requires an unusually complex network architecture and learning mechanism based on known properties of corticothalamic circuits.
Daniel Durstewitz and Jeremy Seamans
Dopaminergic modulation of activity states in network models of the prefrontal cortex
The prefrontal cortex (PFC) is critically involved in working memory, which underlies memory-guided, goal-directed behavior. During working memory tasks, PFC neurons exhibit sustained elevated activity, which may reflect the active holding of goal-related information or the preparation of forthcoming actions. Dopamine via the D1 receptor strongly modulates both this sustained (delay-period) activity and behavioral performance in working memory tasks. However, the function of dopamine during delay-period activity and the underlying neural mechanisms are only poorly understood. Recently we proposed that dopamine might stabilize active neural representations in PFC circuits during tasks involving working memory and render them robust against interfering stimuli and noise. To further test this idea and to examine the dopamine-modulated ionic currents that could give rise to increased stability of neural representations, we developed a network model of the PFC consisting of multi-compartment neurons equipped with Hodgkin-Huxley-like channel kinetics that could reproduce in vitro whole-cell and in vivo recordings from PFC neurons. Dopaminergic effects on intrinsic ionic and synaptic conductances were implemented in the model based on in vitro data. Simulated dopamine strongly enhanced high, delay-type activity but not low, spontaneous activity in the model network. Furthermore, the strength of afferent stimulation needed to disrupt delay-type activity increased with the magnitude of the dopamine-induced shifts in network parameters, making the currently active representation much more stable. Stability could be increased by dopamine-induced enhancements of the persistent Na+ and NMDA conductances. Stability was also enhanced by a reduction in AMPA conductances. The increase in GABAA conductances that occurs following stimulation of dopaminergic D1 receptors was necessary in this context to prevent uncontrolled, spontaneous switches into high-activity states (i.e., spontaneous activation of task-irrelevant representations). In conclusion, the dopamine-induced changes in the biophysical properties of intrinsic ionic and synaptic conductances conjointly acted to highly increase stability of activated representations in PFC networks, and at the same time retain control over network behavior and thus preserve its ability to adequately respond to task-related stimuli.
Scott Makeig, Sigurd Enghoff, Jeanne Townsend (UCSD), Eric Courchesne (UCSD) and Marissa Westerfield (UCSD)
The electrical origins of early event-related potentials
Traditionally, it has been assumed that underlying sources of the electroencephalogram (EEG) and event-related potentials (ERP) are spatially stationary across time. We used infomax Independent Component Analysis (ICA) to test spatial stability in single-trial EEG epochs from ERP experiments. EEG recordings extending 1 sec before a visual stimulus and one second afterwards were analyzed by a 500 ms moving window with successive 14 ms overlap to obtain temporal sequences of independent decompositions. To improve stability of components, a merging scheme was developed which when applied to decomposition of data subsets produced sets of representative components. Results of the moving-window ICA decomposition were comparable to results of decomposing all the single-trial data from each subject at once. However, the moving-window decomposition was more consistent across subjects. Spatially moving components were rare in recordings acquired during a visual selective attention task, but spectrally non-stationary components were common. Time-frequency analysis was applied to the experimental recordings from fifteen subjects. Analyses of event-related changes in spectral coherence revealed an early-event related phase-locking phenomenon, producing the 'alpha-ringing' observed in single-channel ERP averages. These results are consistent with the possibility that phase dynamics, rather than amplitude or energy changes, account for most features of the early visual event-related response.
Kevin Franks and Thomas Bartol
Modeling postsynaptic calcium dynamics
The goal of this study was to determine the Ca2+ entry into postsynaptic spines during conditions that occur during synaptic plasticity. Calcium is an important second messenger in many biological systems. It plays a crucial role in both the activity-induced potentiation and depression of synapses, expressed as LTP and LTD, respectively. These differential changes may be due to a spatio-temporally-selective interaction of Ca2+ with downstream reaction partners. What these are, and the precise spatio-temporal Ca2+ profile is, remains unclear. Using NEURON and MCell, we are simulating the influx of Ca2+ and its interaction with various binding proteins and cytosolic buffers to try to elucidate characteristic potentiating and depressing Ca2+ profiles.
Jean-Marc Fellous, Arthur Houweling, Rashmi Modi, Rajesh Rao, Paul Tiesinga
The frequency dependence of reliability of spike timing differs for pyramidal cells and interneurons
Cortical pyramidal cells and interneurons exhibit subthreshold oscillations when depolarized by current injection. Around threshold, pyramidal and interneuron oscillations are limited to the theta range (4-8 Hz). In the spiking regime, oscillations in pyramidal cells remain in the theta range, but interneuron oscillations enter the gamma range (20-80 Hz). Near spiking threshold, injection of sinusoidal current at the soma of both pyramidal cells and interneurons spike only for a limited range of frequencies, a phenomenon we call 'resonance', in the theta band. The reliability of spike timing was also measured and in the low input regime both pyramidal cells and interneurons are the most reliable when their inputs are in the theta range, consistent with their resonant frequency and intrinsic subthreshold oscillation frequency. In the high input regime, however, pyramidal cells remain reliable in the theta range, while interneuron reliability is the highest in the gamma range. Models of pyramidal cells and interneurons successfully reproduced these phenomena. The effects of acetylcholine and dopamine on resonance and reliability were examined both experimentally and with compartmental models. Information theory is also being used to determine to what extent mutual information in pyramidal cells and interneurons depends on the frequency content of their inputs. In conclusion, pyramidal cells and interneurons have different intrinsic preferences for frequency and synchronicity of synaptic inputs at high levels of activation.
Paul Ganter and Ole Paulsen (Oxford)
The role of morphology for the resonance properties of single cells in the hippocampal CA1 network
Because dendrites have active currents as well as intrinsic passive time constants, resonance and reliability of spiking in response to dendritic current injection may be different than when the current is injected in the soma. We are currently working on detailed multi-compartment models of CA1 pyramids and interneurons taking into account their morphology in parallel with dendritic recordings carried out At Oxford by Ole Paulsen. This should enable us to build reduced models with fewer compartments, which are simple enough to be connected together in a network that includes the spatial organization of the hippocampus. The ultimate aim is to understand the input-output relation and thereby the signal processing properties of the network of pyramidal cells and interneurons as a whole.
Arthur Houweling, Francois Grenier (Laval University, Canada), Igor Timofeev (Laval University, Canada) and Mircea Steriade (Laval University, Canada)
Regulation of spindle oscillations by short-term synaptic plasticity
Sleep spindles consist of waxing and waning 7-14 Hz oscillations that appear in the EEG during the early stages of sleep. Spindle oscillations are generated in the thalamus as a consequence of the intrinsic properties of thalamocortical (TC) and thalamic reticular (RE) cells and their synaptic interactions. Once initiated the waxing of spindles is thought to involve a progressive recruitment of TC cells into the oscillation. In vitro studies suggest that the waning and termination of spindles are caused by a progressive depolarization of TC cells through a calcium-dependent upregulation of the H-current. In vivo intracellular recordings of TC cells, however, rarely display a progressive depolarization during spindles. Recent studies have provided evidence for short-term synaptic plasticity of RE-TC connections. We investigate the consequences of adding short-term plasticity of RE-TC connections in models of spindle oscillations. Preliminary results in an in vitro computational model indicate that the addition of short-term synaptic plasticity does not interfere with the generation of spindle rhythmicity and that the duration of oscillations is reduced. These results suggest that short-term synaptic plasticity of thalamic reticular connections may help terminate spindle oscillations in vivo when the calcium-mediated upregulation of the H-current is weak. Current work addresses the effects of short-term synaptic plasticity of intrareticular and GABAA and GABAB type RE-TC synapses in models of in vivo spindle oscillations.
Tzyy-Ping Jung, Scott Makeig and Sigurd Enghoff
Analysis of single-trial event-related potentials
Most studies of event-related potentials recorded from the scalp average over many trials because of variability in single trials. We have used Independent Component Analysis (ICA) to derive spatial filters that separate single trials into a sum of temporally independent and spatially fixed components arising from distinct or overlapping brain or extra-brain sources. Both the data and their decomposition are displayed using an `ERP image,' that can clearly characterize single-trial variations in the amplitudes and latencies of evoked responses. The new single-trial analysis and visualization tools have been used to analyze data from 28 normal controls and 22 clinical subjects performing a visual selective attention task. Results show that ICA can be used to effectively detect, separate and remove ocular artifacts from EEG recordings. The results compare favorably to those obtained using rejection or regression methods. The ICA method can preserve all of the recorded trials and all of the data channels for response averaging, even if none of the raw trials are artifact-free. Other components derived from the single-trial target response data had a variety of distinct relations to task events. Some captured one category of non-time locked EEG activity, such as blinks, eye-movements, and muscle artifacts. Some were clearly time- and phase-locked to stimulus onsets, while others were time- and phase-locked to subject responses. Still others captured some types of event-modulated oscillatory or other background EEG phenomenon. These analysis and visualization tools enhance the amount and quality of information in event- or response-related brain signals that can be extracted from ERP data, and appear broadly applicable to electrophysiological research on normal and clinical populations.
Tzyy-Ping Jung and Scott Makeig
Eye activity correlates of workload
Several eye activity measures, most notably blink and pupil diameters measures, have been shown to correlate with task workload. Blink rate declines as a function of increased workload. Higher blink rates and longer blink durations have both been associated with task-induced increases in saccadic extent induced by higher workload level. We recently found that blink, fixation and pupil diameter are significantly correlated with workload in a target identification task. Artificial neural network models, derived through training on two or three sessions and subsequently tested on a different session from the same subject, correlated well with actual target density, suggesting that information from multiple eye measures may be combined to produce reliable near-real time indicators of workload in some visuospatial tasks.
Sidney R. Lehky (Riken Brain Science Institute, Japan)
Seeing white: Qualia in the context of decoding population codes
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. This raises 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.
Scott Makeig and Tzyy-Ping Jung
Simultaneous electromagnetic recording and functional magnetic resonance brain imaging
We have used a custom EEG amplifier with timeout circuits and shielded cabling to record single-trial event-related potentials (ERPs) from a healthy adult subject who attended to shapes flashed on a projector screen during continuous 1.5T fMRI scanning. After time-limited artifact rejection, the `ERP-image' showed that the response-locked brain activity was precisely time locked to subject responses in nearly all trials, while the stimulus-locked ERP was time-locked to stimulus onsets. The average of the available ERPs included the well-documented P1/N1/P2/N2 and target LPC components. The functional magnetic resonance imaging (fMRI) data from this experiment showed no decrement in image quality from the EEG electrodes or recording equipment. Standard correlational analysis of the fMRI data showed task-related activations in motor cortex and right anterior cerebellum.
Te-Won Lee and Thomas Wachtler
Learning the chromatic structure of natural scenes
Simple cells in the primary visual cortex (V1) of mammals are selective to bars and edges at different locations, and with different orientations and sizes. Analysis of natural images using unsupervised learning algorithms have suggested that these neurons form a sparse representation that efficiently encodes natural scenes. We have applied independent component analysis (ICA) to hyperspectral images of chromatic natural images. In the spectra of single pixels, the algorithm found basis functions that had broadband spectra similar to natural daylight, as well as basis functions that coincided with the human cone spectral sensitivity functions. When applied to small image patches, the algorithm found achromatic basis functions and basis functions with overall chromatic variation along lines in color space, consistent with color opponency. This suggests that the brain has evolved neural systems that efficiently represent the underlying chromatic structure of natural scenes.
Te-Won Lee and Michael S. Lewicki
Unsupervised classification, segmentation and de-noising of images using ICA mixture models
We applied a probabilistic method for learning efficient image codes to the problem of unsupervised classification, segmentation and de-noising of images. The method is based on the Independent Component Analysis (ICA) mixture model proposed for unsupervised classification and automatic context switching in blind source separation. We demonstrate that this algorithm is effective in classifying complex image textures such as trees and rocks in natural scenes. The algorithm is useful for de-noising and filling in missing pixels in images with complex structures. The advantage of this model is that image codes can be learned with increasing numbers of basis function classes. Our results suggest that the ICA mixture model provides greater flexibility in modeling structure and in finding more image features than in either Gaussian mixture models or standard ICA algorithms.
Daniel Needleman and Paul Tiesinga
Collective enhancement of precision in networks of coupled oscillators
We have analyzed conditions under which the precision of coupled oscillators can be improved by synchronization. The improvement in precision, defined as the inverse of the coefficient of variation of the periods, depends on how noise is added to the system. If the magnitude of noise experienced by an oscillator only depends on the state of that oscillator, then the precision of the group rhythm can be improved to 1/sqrt(N) times the precision of an individual uncoupled oscillator, irrespective of the form of the coupling function. However, if the magnitude of the noise also depends on the state of other oscillators, as might be the case for noise caused by synaptic input to neurons, then synchronization may lead to an additional improvement. This collective enhancement of precision is demonstrated both with a simple phase model and a network of integrate-and-fire neurons.
Daniel Needleman and Amelia Stanco
Reversible temperature induced desynchronization of the pacemaker nucleus of a weakly electric fish
The pacemaker nucleus of the weakly electric fish Apteronotus leptorhynchus is a network of approximately one hundred and fifty neurons that control the electric organ discharge on a beat-by-beat basis. The neurons in the pacemaker nucleus are synchronized to a common frequencies that is maintained with a remarkable precision. We have studied the effect of controlled temperature changes on the isolated pacemaker nucleus and the intact fish. Gradual increases in temperature raises the frequency of oscillation but does not effect the neurons precision. However, sudden increases in temperature can cause the nucleus to desynchronize. This effect is reversible, and the nucleus could be repeatedly brought into and out of synchronization by adjusting the temperature. A closer study of this phenomena may help to elucidate how the interactions between neurons may cause a population of neurons to synchronize to a common frequency. More specifically, this relatively simple system might allow mathematical theories of synchronization to be tested.
Michael Neubig
What are the spatio-temporal characteristics of dendrites?
How do arbor geometries facilitate or limit it dendritic integration of synaptic signals? What characteristics of the "boundary" conditions distinguish propagating events from whole-cell events? When whole-cell events dominate, and this intermediate scale of organization collapses, what are the implications from a dynamical systems perspective? Compartmental models of reconstructed neurons and modified reconstructions are being used to address these questions.
K.T. Moortgat and Theodore H. Bullock (UCSD)
Network and cellular influences on temporal precision in the pacemaker nucleus of a weakly electric fish
We studied the relative influence of cellular and network properties on the extreme spike timing precision observed in the medullary pacemaker nucleus of the weakly electric fish Apteronotus leptorhynchus. Of all known biological rhythms, the electric organ discharge of this and related species is the most temporally precise, with a coefficient of variation of 0.0002 and standard deviation of 0.12-1.0 microsecond. From neural responses to intracellular current injections, we estimated the strength of gap junctions between neurons in the nucleus. We found that, when these gap junctions were blocked by pharmacological agents, the cellular precision was largely maintained. Therefore we concluded that the extremely high spiking precision observed in this nucleus is largely due to cellular properties rather than network connectivity. We also investigated the precision of spike timing in a computer model of gap junction-coupled oscillatory neurons. The model incorporated the known physiology, including the morphology and connectivity, of the weakly electric fish's high frequency and extremely precise pacemaker nucleus (Pn). Each neuron were modeled with two compartments containing Hodgkin-Huxley sodium and potassium currents. When coupled by gap junctions to other neurons, a model neuron, like its biological correlate, spiked at frequencies and amplitudes that were largely independent of current injections. Gap junctions facilitate synchronization and can improve temporal precision, are most effective between axons. We conclude that cells in the network must have high intrinsic precision to account for the precision of cells observed in the biological network.
Rajesh P.N. Rao, Margaret S. Livingstone (Harvard Medical School),
Predictive sequence learning in neocortical circuits and directional selectivity of neurons in visual cortex of primates
Neocortical circuits are dominated by massive excitatory feedback: more than eighty percent of the synapses made by excitatory cortical neurons are onto other excitatory cortical neurons. Why is there such massive recurrent excitation in the neocortex and what is its role in cortical computation? Previous modeling studies have suggested a role for excitatory feedback in amplifying feedforward inputs. Recently, it has been shown that recurrent excitatory connections between cortical neurons are modified according to a temporally asymmetric Hebbian learning rule: synapses that are activated slightly before the cell fires are strengthened whereas those that are activated slightly after are weakened. In biophysical simulations, a population of recurrently connected neurons with this form of synaptic plasticity can learn to predict spatiotemporal input patterns. In particular, we have demonstrated that a network of cortical neurons can develop direction selectivity similar to that observed in complex cells in alert monkey visual cortex as a consequence of learning to predict moving stimuli.
Philip Sabes and Richard Andersen (Caltech)
Reference frames in parietal cortex
We have been investigating the neural coding of eye movements in primate parietal cortex. The lateral intraparietal area (LIP) is responsible for the early planning of saccadic eye movements. While there has been a good deal of work exploring how LIP encodes movement plans, the bulk of that work has studied movements to points of light on a homogeneous background. We set out to study how interaction with an object effects the planning of eye movements. Monkeys were trained to perform a novel object-based saccade task: a cue is flashed on an object, and after the object rotates in space, the animal must saccade to the location on the object where the cue had been. We found that LIP neurons are primarily tuned for the retinotopic direction of movement or the orientation of the object on the screen. No cells primarily code for the location of the movement in an object-fixed coordinate frame. Yet when we rotate and translate the object, many cells show a strong tuning for the direction of movement with respect to the center of the object, irrespective of the object's orientation in space, consistent with an attention-centered tuning. The object-based saccade task also allowed us to address how the cortex forms stable representations of a dynamic environment for guiding the motor system. As the object rotates in space, the required movement plan changes. Using novel analytic techniques, we explored how the representations for movement in LIP evolve during this change in plan. The dynamics of the evolution depended critically on the visual feedback available to the animal. When the object visibly rotates in front of the animal, the population code shows a smooth rotation from the cue direction to the movement direction even though the cue is no longer visible on the object, as if the area were performing ``attentional pursuit'' of the relevant location on the object. On the other hand, if the object blanked and reappeared at a new orientation, the neural representation shows a discrete shift from one direction to the other. Furthermore, the timing of that shift correlated across trials with the animal's reaction time rather than the time of stimulus onset, offering further evidence that LIP plays a key role in transforming visual input into motor commands.
Emilio Salinas
How correlated input alters the responses of cortical neurons
Neurons in the cortex are typically driven by thousands of synaptic inputs. A postsynaptic neuron integrates those inputs and responds by producing a spike train, whose statistics depend on the statistics of the inputs. The statistics of the output spike train are usually characterized by the mean number of spikes emitted per unit time, the firing rate, and by the variance of the interspike interval distribution divided by the mean of that distribution, known as the coefficient of variation, or CV. Numerous experimental and theoretical studies have indicated that a postsynaptic neuron is sensitive to correlations between inputs. That is, the output mean rate and CV are different when excitatory and inhibitory spikes arrive independently (uncorrelated inputs) and when the arrival of a spike from one input reflects an increase or a decrease in the chances that another spike from a different input arrives closely in time (correlated inputs). This dependence is crucial to normal cortical function, because, since neurons are embedded in interconnected microcircuits, their inputs are not independent, they may exhibit various degrees of correlation. However, a clear understanding of how the statistics of the output neuron depend on the statistics of its inputs has been lacking. A stochastic model was developed for which an analytical solution for the output rate can be calculated; the output firing rate is expressed as a function of the mean rates and pairwise correlations of the inputs. According to the model, correlations may have a variety of effects on the rate, depending on model parameters and on the relative weights of correlations between excitatory-excitatory, inhibitory-inhibitory and excitatory-inhibitory pairs of neurons. Using biologically plausible parameters, the model predicts that correlations between excitatory inputs should increase the firing rate of the target neuron, particularly around threshold. A more realistic, conductance-based model implemented through computer simulations has also been developed in order to confirm and extend these results.
Laura Schultz
Low-threshold calcium channels in CA1 hippocampal interneurons
In the hippocampus, low-threshold (T-type) calcium currents have been implicated in subthreshold membrane fluctuations, burst firing, rhythmic oscillations, and resonance in a variety of neurons. The spatial distribution of T-type calcium channels in stratum oriens and stratum radiatum interneurons in area CA1 affects their responses to synaptic inputs at different locations in the dendritic tree. We were interested in determining whether these interneurons possess T-type channels. Using a combination of whole-cell patch recording and confocal microscopy, we found that stratum oriens interneurons possess nickel-sensitive T-type channels in both their soma and dendrites. By contrast, T-type channels appear to be concentrated in the dendrites of stratum radiatum interneurons. Interestingly, our results suggest that the T-type channels in these interneurons may be modulated by glutamate. Neither class of interneuron had a significant amount of pimozide-sensitive.
Jeremy Seamans and Daniel Durstewitz
Analysis of prefrontal cortical function
The prefrontal cortex (PFC) is a region of the brain involved in that uses of contextual information and working memory to plan and organize forthcoming behaviors. The goal of this research is to build a detailed theoretical model of the PFC from the cellular level, which could explain how this region of the brain processes information in normal and pathological states. Electrophysiological recordings from prefrontal slices of cortex have revealed that dopamine D1 receptors produce multiple effects on PFC pyramidal neurons including an increase in the persistent Na+ current and NMDA currents as well as a slight decrease in pre-synaptic glutamate release. Concurrently dopamine D1 receptors act on a specific subclass of inhibitory interneurons to increase their excitability and augment their ability to inhibit pyramidal cells. We used small-interconnected networks of realistic PFC neurons to understand the functional consequences of these cellular effects of dopamine. The models indicate that the dopamine-induced changes decreased spontaneous firing of pyramidal neurons but enhanced the connectivity of neighboring deep layer pyramidal neurons by producing sustained steady-state depolarizations to synaptic input trains. This favored recurrent excitation of the type thought to underlie the active retention of information by pyramidal cells in the PFC.
Roland E. Suri
Computational function of sequence compression within hippocampal theta cycles
Neural activity of a subset of hippocampal neurons reflects the rat's location in a known environment (place cells). This neural activity is modulated at theta frequencies (4-12 Hz) and when a rat walks through the place field of such a neuron, the activity progressively shifts to earlier phases of the theta rhythm, which is called phase precession. Therefore, the activity of neurons with overlapping place fields repeats the temporal sequence of hippocampal place field activations within each theta cycle in compressed form. Although several models reproduce these experimental findings, the function of this sequence compression is still unclear. A model developed previously for the cortex, basal ganglia, and dopamine neurons reproduced anticipatory neural activities by assuming that expected future sequences are compressed in each cycle of the rhythmic theta activity. In this model, the time compression mechanism emulated the anticipated sensory input. Preliminary work suggests that hippocampal sequence compression could be related to internal models that appear in engineering control algorithms.
Paul Tiesinga, Jean-Marc Fellous and Jorge V. Jose (Northeastern)
Information transmission by neocortical cells
We studied the output precision and information transmission capacity of a model neocortical neuron and prefrontal cortex pyramidal cells in slice preparation, driven by a finite precision periodic (40 Hz) presynaptic spike train. Spike-timing precision is maintained during feed-forward propagation either with weak synapses during entrainment, or with strong inhibitory synapses outside entrainment. The required number of presynaptic inputs and their precision were determined to insure that the entrainment is stable. During entrainment the Shannon information capacity of the output is reduced. However, paradoxically, the amount of transmitted information is increased. Furthermore, part of the information content residing in the precise presynaptic spike times is transmitted. We also quantified the robustness of the information transmission in the presence of intrinsic neuronal noise sources. We propose that neuromodulation, via the nonlinear dynamics entrainment, can quantitatively modulate the information transfer in neocortical networks and saliency of stimuli.
Paul Tiesinga, Jean-Marc Fellous, Jorge V. Jose (Northeastern) and Shuang Zhang.
Regulation of rhythmogenesis in the hippocampus
Field potentials from the rat hippocampus have revealed a wide range of distinct frequency bands of activity, including delta (0.5-2 Hz), theta (4-12 Hz), and gamma (30-80 Hz) that depend on the behavioral state. Application of the carbachol (CCH), a cholinergic agonist, to a hippocampal slice induces oscillations in the delta (CCH-delta), theta (CCH-theta), and gamma (CCH-gamma) frequency range, depending on the concentration. As the concentration of carbachol was increased, incoherent CCH-theta, synchronous CCH-delta, and synchronous CCH-theta emerged sequentially. In a model network of CA3 cells, the time scale for CCH-delta corresponded to the decay constant of the gating variable of the calcium-dependent potassium current (K-AHP), the CCH-theta was derived from an intrinsic subthreshold membrane potential oscillation in the pyramidal cells, and the CCH-gamma was due to the decay constant of GABA-ergic inhibitory synaptic potentials on pyramidal cells. Simulations have shown how the known physiological effects of carbachol on the muscarinic and K-AHP currents and the distribution of the strength of excitatory postsynaptic potentials could account for the transitions from incoherent CCH-theta to CCH-delta and from CCH-delta to synchronous CCH-theta. The simulations also exhibited the nested CCH-gamma-CCH-delta and CCH-gamma-CCH-theta observed in experiments. A new dynamical state was predicted in which all three frequency bands coexist.
Thomas Wachtler and Thomas Albright (Salk Institute)
Cortical coding and perception of color
Although color vision is among the oldest and best-studied part of vision, the neural processes underlying color perception are largely unknown. The goal of this study was to investigate how stimuli that are commonly used in psychophysical experiments on color perception are represented in primary visual cortex. When tested with color stimuli presented on colored backgrounds, many neurons in V1 showed color selectivity, responding better to certain colors than to others. The colors that elicited the largest responses were not equally distributed in color space, suggesting that the cortical representation of colors is based on a population code with certain preferred colors. In almost all neurons, responses were influenced by the color of the background. This influence was often very specific and affected only the responses to colors similar to the background color. Comparison of the neural responses with perceptual judgments of the stimuli showed that the magnitudes of the respective effects of background colors were comparable. The findings suggest that the representation of colors in primary visual cortex carries important aspects of the final color percept.
Martina Wicklein
Active visual depth perception by looming
Looming-sensitive neurons have been found in Manduca sexta that signal changes in depth, but the cues in the visual scenes that drive these cells are unknown. Models of the visual responses related to looming detection are being developed and experimentally tested with electrophysiological recordings. Perception of change in depth is a key feature for visual orientation, for moving through space, approaching objects, avoiding collisions, initiating escape behavior, or stabilizing a position at a given distance from a moving object. Intracellular recordings in Manduca sexta (Sphingidae, Lepidoptera) have identified wide-field neurons that responded to looming or receding stimuli. Looming and anti-looming conditions were simulated by four different stimuli: a spiral rotating clockwise- and anti-clockwise and an expanding or contracting disc. Both spiral and disc provide the eye with outwardly or inwardly moving edges. Two types of wide-field visual neurons were found in the optic lobes of M. sexta that are selectively activated by these parameters. Type 1 cells discriminate an approaching or receding disc from an outwardly or inwardly rotating spiral, being activated only by the disc and not by the spiral and do not respond to moving gratings. Thus, type 1 cells respond to parameters of the image other than motion stimuli, distinguish expansion from contraction on the base of changing perimeter length. Type 2 neurons are excited both by movement of the spiral and by an approaching or receding disc and also respond vigorously to horizontal and vertical moving gratings. Type 2 cells use flowfield patterns to distinguish between expanding and contracting objects. Computational network models based on the characteristics of type1 and type 2 neurons are being developed. The models are tested with the same stimuli used for stimulating the neurons and in addition with a host of new stimuli to find critical stimuli that will subsequently be used to test the neurons. Stimuli will be constructed that can be used to discriminate between these different models and will be used in further physiological experiments.
Kechen Zhang
A universal scaling law between gray matter and white matter in cerebral cortex
The cerebral cortex of mammals is a relatively recently evolved brain structure and that has a similar layered architecture in species over a wide range of brain sizes. Long-distance communication between different cortical areas becomes more demanding in larger brains: The volume of the white matter containing long axons increases disproportionately faster than the volume of the gray matter that contains cell bodies, dendrites, and axons for local information processing, according to a power law. The theoretical analysis presented here shows how this remarkable anatomical regularity might arise naturally as a consequence of the local uniformity of the cortex and the requirement for compact organization of long axonal fibers. The predicted power law with an exponent of 4/3 minus a small correction for the thickness of the cortex accurately accounts for empirical data spanning 5 to 6 orders of magnitude in brain sizes for various mammalian species including human and other primates.
Kechen Zhang
Stabilizing a sequence of point attractors
A dimensionality reduction method is presented to accurately characterize the transient dynamics of a general class of asymmetric attractor networks that can store sequences of memory patterns using only Hebbian-type connection weights. Although an asymmetric network in general does not admit a Lyapunov function, its dynamics can be folded and compressed into a self-contained deterministic system, whose emergent oscillatory behavior predicts the stability as well as the speed of the spontaneous sequential retrieval in the original network. The reduction procedure can be summarized by a few reduction rules, which are derived and applied to several solvable networks, with the analytical solutions confirmed by numerical simulations. This approach provides new quantitative tools for analyzing the nonlinear dynamics of some common asymmetric networks for the storage and retrieval of sequential memory patterns, including the time delay networks and the coupled Hopfield networks as special cases.
Publications 1999
Articles
Zhang, K., and Sejnowski, T., J., Neuronal tuning: To sharpen or broaden?, Neural Computation 11(1), 75-84, (1999).
Bazhenov, M., Timofeev, I., Steriade, M., and Sejnowski, T. J., Self-sustained rhythmic activity in the thalamic reticular nucleus mediated by GABAA potentials, Nature Neuroscience 2(2), 168-174 (1999).
Lee, T.-W., Girolami, M., and Sejnowski, T. J., Independent component analysis using an extended infomax algorithm for mixed Subgaussian and Supergaussian sources, Neural Computation 11(2), 417-441 (1999).
Makeig, S., Westerfield, M., Jung, T.-P., Covington, J., Townsend, J., Sejnowski, T. J., and Courchesne, E., Functionally independent components of the late positive event-related potential during visual spatial attention, Journal of Neuroscience 19(7) 2665-2680 (1999).
Stewart-Bartlett, M., Hager, J., Ekman, P., and Sejnowski, T. J., Measuring facial expressions by computer image analysis, Psychophysiology, 36(2) 253-263 (1999).
Lee, T.-W., Lewicki, M. S., Girolami, M. and Sejnowski T. J., Blind source separation of more sources than mixtures using overcomplete representations, IEEE Signal Processing Letters 6(4) 87-90 (1999).
Zhang, K., and Sejnowski, T. J., A theory of geometric constraints on neural activity for natural three-dimensional movement,Journal of Neuroscience 19(8) 3122-3145 (1999).
Lehky, S., and Sejnowski, T. J., Seeing white: Qualia in the context of decoding population codes, Neural Computation 11(6), 1261-1280 (1999).
Houweling, A. R., Bazhenov, M., Timofeev, I., Steriade, M., and Sejnowski, T. J., Cortical and thalamic components of augmenting responses: A modeling study, Neuralcomputing 26-27, 735-742 (1999).
van Praag, H., Christie, B. R., Sejnowski, T. J., and Gage, F. H., Running enhances long-term potentiation in mice, Proceedings of the National Association of Sciences 96 (23) 13427-13431 (1999).
Makeig, S., Westerfield, M., Townsend, J., Jung, T.-P., Courchesne, E., and Sejnowski, T. J., Functionally independent components of early event-related potentials in a visual spatial attention task, Philosophical Transactions: Biological Sciences-The Royal Society 354, 1135-44, 1999.
Ernst, U., Pawelzik, K., Tsodyks, M., and Sejnowski, T. J., Relation between retinotopic and orientation maps in visual cortex, Neural Computation 11(2), 375-379 (1999).
Destexhe, A., McCormick, D. A., and Sejnowski T. J., Thalamic and thalamocortical mechanisms underlying 3 Hz spike-and-wave discharges, Prog Brain Res., 121, 81-97(1999).
Invited Reviews
Sejnowski, T., J., Computational Neuroscience, The MIT Encyclopedia of the Cognitive Sciences, Robert A. Wilson and Frank Keil (Eds.), MIT Press, 165-168, (1999).
Sejnowski, T. J., A high point for evolution, Science 283, 1121 (1999).
Sejnowski, T. J., Computational neuroscience, Encyclopedia of Neuroscience, Amsterdam: Elsevier, 450-453 (1999).
Sejnowski, T. J., Neural networks beyond Freud, Nature 400, 632-633 (1999).
Sejnowski, T. J., The Book of Hebb, Neuron 24, 1-20, (1999).
Edited Books
Abbott, L. F. and Sejnowski, T. J., Neural Codes and Distributed Representations: Foundations of Neural Computation. L. F. MIT Press, Cambridge MA, 1999.
Hinton, G. E. and Sejnowski, T. J., Unsupervised Learning: Foundations of Neural Computation, MIT Press, Cambridge, MA, 1999.
Refereed Conference Proceedings
Lee, T.-W., Lewicki, M. S., and Sejnowski, T. J., ICA mixture models for unsupervised classification and automatic context switching, International workshop on Independent component analysis and blind signal separation, .January 11-15, Aussois, France, 209-214 (1999).
Jung, T.-P., Makeig, S., Westerfield, M., Townsend, J., Courchesne, E., and Sejnowski, T. J., Independent component analysis of single-trial event-related potentials, International workshop on Independent component analysis and blind signal separation, .January 11-15, Aussois, France, 173-178 (1999).
Jung, T.-P., Makeig, S., Westerfield, M., Townsend, J., Courchesne, E., and Sejnowski, T. J., Analyzing and visualizing single-trial event-related potentials, Advances in Neural Information Processing Systems 11, 118-124, 1999.
Lee, T.-W., Lewicki, M. S., and Sejnowski, T. J., Unsupervised classification with non-gaussian mixture models using ICA, Advances in Neural Information Processing Systems 11, 508-514, 1999.
Lewicki, M..S., and Sejnowski, T..J., Coding time-varying signals using sparse, shift-invariant representations, Advances in Neural and Information Processing Systems 11, 730-736, 1999.
Lee, T. -W., Lewicki, M. S., and Sejnowski, T. J., ICA mixture models for unsupervised classification and automatic context switching, International workshop on Independent component analysis and blind signal separation, January 11-15, 1999, Aussois, France, 209-214, Jan.1999.
Ernst, U., Pawelzik, K., Tsodyks, M., Sejnowski, T., and Geisel, T., Relationships between cortical maps and receptive fields are determined by lateral cortical feedback, Proceedings of the 27th Goettingen Neurobiology Conference, Georg Theime Verlag Stuttgart, New York (1999).
Lee, T.-W., Lewicki, M. S., and Sejnowski, T. J., ICA mixture models for image processing, Proceedings of the 6th Joint Symposium on Neural Computation, Institute for Neural Computation, UCSD, 79-86, (1999).
Stewart-Bartlett, M., Donato, G., Movellan, J. R., Hager, J. C., Ekman, P., and Sejnowski, T. J., Face image analysis for expression measurement and detection of deceit, Proceedings of the 6th Joint Symposium on Neural Computation, Institute for Neural Computation, UCSD, 8-15 (1999).
Kreutz-Delgado, K., Rao, B. D., Engan, K., Lee, T. -W., and Sejnowski, T. J., Convex/Schur-convex (CSC) log-priors and sparse coding, Proceedings of the 6th Joint Symposium on Neural Computation, Institute for Neural Computation, UCSD, 65-71, (1999).
Kreutz-Delgado, K., Rao, B. D., Engan, K., Lee, T.-w., and Sejnowski, T., Learning overcomplete dictionaries: Convex/Schur-convex (SCS) log-priors, factorial codes, and independent/ dependent component analysis (I/DCA), Proceedings of the 6th Joint Symposium on Neural Computation, Institute for Neural Computation, UCSD, 72-78, (1999).
Bazhenov, M., Stopfer, M., Rabinovich, M., Abarbanel, H. D. I., Sejnowski, T. J., and Laurent, G., Network model for the odor-specific temporal patterns in locust olfactory interneurons, Proceedings of the 6th Joint Symposium on Neural Computation, Institute for Neural Computation, UCSD, 16-23, (1999).
Rao, R. P. N., and Sejnowski, T. J., Direction selectivity from predictive sequence learning in recurrent neocortical circuits, Proceedings of the 6th Joint Symposium on Neural Computation, Institute for Neural Computation, UCSD, 119-126, (1999).
Seamans, J. K., Durstewitz, D., and Sejnowski, T. J., State-dependence of dopamine D1 receptor modulation in prefrontal cortex neurons, Proceedings of the 6th Joint Symposium on Neural Computation, Institute for Neural Computation, UCSD, 128-135, (1999).
Book Chapters
Destexhe, A., McCormick, D. A., and Sejnowski T. J., Thalamic and thalamocortical mechanisms underlying 3 Hz spike-and-wave discharges, In: J. Reggia, E. Ruppin, and D. Glanzman (Eds.) Brain, Behavioral, and Cognitive Disorders: The Neurocomputational Perspective, Elsevier, Amsterdam, 289-307 (1999).
Abbott, L. F., and Sejnowski, T. J., Neural codes and distributed representations: Foundations of neural computation, In: L. F. Abbott and T. J. Sejnowski (Eds.) Neural codes and distributed representations: Foundations of neural computation, MIT Press, Cambridge, MA, vii-xxiii (1999).
Doya, K., and Sejnowski, T. J., A computational model of avian song learning, In: M. S. Gazzaniga (Ed.), The New Cognitive Neurosciences, 2nd edition, pages 469-482, MIT Press, Cambridge, MA (1999).
Pouget, A., Deneve, S. and Sejnowski, T. J., Frames of reference in hemineglect: A computational approach, In: J. Reggia, E. Ruppin, and D. Glanzman (Eds.) Progress in Brain Research: Brain, Behavioral, and Cognitive Disorders: The Neurocomputational Perspective, Elsevier, Amsterdam, 121, 81-97 (1999).
Abstracts
Schultz, L. M., Christie, B. R., and Sejnowski, T. J., Distribution of T-type calcium channels in CA1 stratum oriens interneurons, Society for Neuroscience Abstract 25, 198 (1999).
Egelman and Sejnowski, T. J., A flashed stimulus perceptually lags a moving one due to rewriting of the past, not extrapolation into the future, Society for Neuroscience Abstract 25, 399 (1999).
Fellous, J.-M., Rao, R. P. N., Houweling, A. R., and Sejnowski, T. J., Spike timing reliability in the prefrontal cortex depends on the frequency content of its synaptic inputs, Society for Neuroscience Abstract 25, 885 (1999).
van Praag, H., Christie, B. R., Sejnowski, T. J., and Gage, F. H., Running enhances neurogenesis, learning and long-term potentiation (LTP) in mice, Society for Neuroscience Abstract 25, 888 (1999).
Coenen, O. J.-M.D., Arnold, M., Jabri, M.A., Courchesne, E., and Sejnowski, T. J., A hypothesis for parallel fiber coding in the cerebellum, Society for Neuroscience Abstract 25, 915 (1999).
Durstewitz, D., Seamans, J., K., and Sejnowski, T. J., Dopaminergic modulation of activity states in the prefrontal cortex, Society for Neuroscience Abstract 25, 1216 (1999).
Rao, R. P. N., Livingstone, M. S., and Sejnowski, T. J., Direction selectivity from predictive sequence learning in recurrent neocortical circuits, Society for Neuroscience Abstract 25, 1316 (1999).
Bazhenov, M., Timofeev, I., Steriade, M., and Sejnowski, T. J., Model of spontaneous activity in the isolated cortical slab in vivo, Society for Neuroscience Abstract 25, 1660 (1999).
Westerfield, M., Makeig, S., Townsend, J., Jung, T. -P., and Sejnowski, T. J., Functionally independent components of early visual event-related potentials, Society for Neuroscience Abstract 25, 1631 (1999).
Makeig, S., Townsend, J., Jung, T.-P., Enghoff, S., Gibson, C., and Sejnowski, T. J., Early visual evoked response peaks appear to be sums of activity in multiple alpha sources, Society for Neuroscience Abstract 25, 1631 (1999).
Franks, K. M., Bartol, T. M., Egelman, D. M., Poo, M. M. and Sejnowski, T. J., Adaptation of CA2 homeostasis in CA1 pyramidal neurons of calbindin-D28K knockout mice, Society for Neuroscience Abstract 25, 1989 (1999).
Chan, K., Zhang, K., Knierim, J. J., McNaughton, B. L. and Sejnowski, T. J., Comparison of different methods for position reconstruction from hippocampal place cell recordings, Society for Neuroscience Abstract 25, 2166 (1999).
Jung, T. -P., Makeig, S., Townsend, J., Westerfield, M., Hicks, B., Courchesne, E., and Sejnowski, T. J., Single-trial ERPS during continuous fMRI scanning, Society for Neuroscience Abstract 25, 1389 (1999).
Wachtler, T., Sejnowski, T. J., and Albright, T. D., Responses of cells in macaque V1 to chromatic stimuli are compatible with human color constancy, Society for Neuroscience Abstract 25, 4 (1999).
Houweling, T. R., Grenier, F., Timofeev, I., Steriade, M., and Sejnowski, T. J., Termination of spindle oscillations by short-term synaptic plasticity of thalamic reticular connections, Society for Neuroscience Abstract 25, 361 (1999).
Christie, B. R., and Sejnowski, T. J., NMDA-receptor dependent synaptic plasticity in rat hippocampal stratum radiatum interneurons and giant cells, Society for Neuroscience Abstract 25, 462 (1999).