STDP-Compatible Approximation of Backpropagation in an Energy-Based Model.

Abstract:

:We show that Langevin Markov chain Monte Carlo inference in an energy-based model with latent variables has the property that the early steps of inference, starting from a stationary point, correspond to propagating error gradients into internal layers, similar to backpropagation. The backpropagated error is with respect to output units that have received an outside driving force pushing them away from the stationary point. Backpropagated error gradients correspond to temporal derivatives with respect to the activation of hidden units. These lead to a weight update proportional to the product of the presynaptic firing rate and the temporal rate of change of the postsynaptic firing rate. Simulations and a theoretical argument suggest that this rate-based update rule is consistent with those associated with spike-timing-dependent plasticity. The ideas presented in this article could be an element of a theory for explaining how brains perform credit assignment in deep hierarchies as efficiently as backpropagation does, with neural computation corresponding to both approximate inference in continuous-valued latent variables and error backpropagation, at the same time.

journal_name

Neural Comput

journal_title

Neural computation

authors

Bengio Y,Mesnard T,Fischer A,Zhang S,Wu Y

doi

10.1162/NECO_a_00934

subject

Has Abstract

pub_date

2017-03-01 00:00:00

pages

555-577

issue

3

eissn

0899-7667

issn

1530-888X

journal_volume

29

pub_type

杂志文章
  • Derivatives of logarithmic stationary distributions for policy gradient reinforcement learning.

    abstract::Most conventional policy gradient reinforcement learning (PGRL) algorithms neglect (or do not explicitly make use of) a term in the average reward gradient with respect to the policy parameter. That term involves the derivative of the stationary state distribution that corresponds to the sensitivity of its distributio...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.2009.12-08-922

    authors: Morimura T,Uchibe E,Yoshimoto J,Peters J,Doya K

    更新日期:2010-02-01 00:00:00

  • Changes in GABAB modulation during a theta cycle may be analogous to the fall of temperature during annealing.

    abstract::Changes in GABA modulation may underlie experimentally observed changes in the strength of synaptic transmission at different phases of the theta rhythm (Wyble, Linster, & Hasselmo, 1997). Analysis demonstrates that these changes improve sequence disambiguation by a neural network model of CA3. We show that in the fra...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/089976698300017539

    authors: Sohal VS,Hasselmo ME

    更新日期:1998-05-15 00:00:00

  • Solution methods for a new class of simple model neurons.

    abstract::Izhikevich (2003) proposed a new canonical neuron model of spike generation. The model was surprisingly simple yet able to accurately replicate the firing patterns of different types of cortical cell. Here, we derive a solution method that allows efficient simulation of the model. ...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.2007.19.12.3216

    authors: Humphries MD,Gurney K

    更新日期:2007-12-01 00:00:00

  • Computing confidence intervals for point process models.

    abstract::Characterizing neural spiking activity as a function of intrinsic and extrinsic factors is important in neuroscience. Point process models are valuable for capturing such information; however, the process of fully applying these models is not always obvious. A complete model application has four broad steps: specifica...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00198

    authors: Sarma SV,Nguyen DP,Czanner G,Wirth S,Wilson MA,Suzuki W,Brown EN

    更新日期:2011-11-01 00:00:00

  • Hybrid integrate-and-fire model of a bursting neuron.

    abstract::We present a reduction of a Hodgkin-Huxley (HH)--style bursting model to a hybridized integrate-and-fire (IF) formalism based on a thorough bifurcation analysis of the neuron's dynamics. The model incorporates HH--style equations to evolve the subthreshold currents and includes IF mechanisms to characterize spike even...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/089976603322518768

    authors: Breen BJ,Gerken WC,Butera RJ Jr

    更新日期:2003-12-01 00:00:00

  • Convergence of the IRWLS Procedure to the Support Vector Machine Solution.

    abstract::An iterative reweighted least squares (IRWLS) procedure recently proposed is shown to converge to the support vector machine solution. The convergence to a stationary point is ensured by modifying the original IRWLS procedure. ...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/0899766052530875

    authors: Pérez-Cruz F,Bousoño-Calzón C,Artés-Rodríguez A

    更新日期:2005-01-01 00:00:00

  • Nonmonotonic generalization bias of Gaussian mixture models.

    abstract::Theories of learning and generalization hold that the generalization bias, defined as the difference between the training error and the generalization error, increases on average with the number of adaptive parameters. This article, however, shows that this general tendency is violated for a gaussian mixture model. Fo...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/089976600300015439

    authors: Akaho S,Kappen HJ

    更新日期:2000-06-01 00:00:00

  • A first-order nonhomogeneous Markov model for the response of spiking neurons stimulated by small phase-continuous signals.

    abstract::We present a first-order nonhomogeneous Markov model for the interspike-interval density of a continuously stimulated spiking neuron. The model allows the conditional interspike-interval density and the stationary interspike-interval density to be expressed as products of two separate functions, one of which describes...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.2009.06-07-548

    authors: Tapson J,Jin C,van Schaik A,Etienne-Cummings R

    更新日期:2009-06-01 00:00:00

  • Information loss in an optimal maximum likelihood decoding.

    abstract::The mutual information between a set of stimuli and the elicited neural responses is compared to the corresponding decoded information. The decoding procedure is presented as an artificial distortion of the joint probabilities between stimuli and responses. The information loss is quantified. Whenever the probabilitie...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/089976602317318947

    authors: Samengo I

    更新日期:2002-04-01 00:00:00

  • On the performance of voltage stepping for the simulation of adaptive, nonlinear integrate-and-fire neuronal networks.

    abstract::In traditional event-driven strategies, spike timings are analytically given or calculated with arbitrary precision (up to machine precision). Exact computation is possible only for simplified neuron models, mainly the leaky integrate-and-fire model. In a recent paper, Zheng, Tonnelier, and Martinez (2009) introduced ...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00112

    authors: Kaabi MG,Tonnelier A,Martinez D

    更新日期:2011-05-01 00:00:00

  • Including long-range dependence in integrate-and-fire models of the high interspike-interval variability of cortical neurons.

    abstract::Many different types of integrate-and-fire models have been designed in order to explain how it is possible for a cortical neuron to integrate over many independent inputs while still producing highly variable spike trains. Within this context, the variability of spike trains has been almost exclusively measured using...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/0899766041732413

    authors: Jackson BS

    更新日期:2004-10-01 00:00:00

  • Modeling slowly bursting neurons via calcium store and voltage-independent calcium current.

    abstract::Recent experiments indicate that the calcium store (e.g., endoplasmic reticulum) is involved in electrical bursting and [Ca2+]i oscillation in bursting neuronal cells. In this paper, we formulate a mathematical model for bursting neurons, which includes Ca2+ in the intracellular Ca2+ stores and a voltage-independent c...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.1996.8.5.951

    authors: Chay TR

    更新日期:1996-07-01 00:00:00

  • Alignment of coexisting cortical maps in a motor control model.

    abstract::How do multiple feature maps that coexist in the same region of cerebral cortex align with each other? We hypothesize that such alignment is governed by temporal correlations: features in one map that are temporally correlated with those in another come to occupy the same spatial locations in cortex over time. To exam...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.1996.8.4.731

    authors: Chen Y,Reggia JA

    更新日期:1996-05-15 00:00:00

  • Statistical computer model analysis of the reciprocal and recurrent inhibitions of the Ia-EPSP in α-motoneurons.

    abstract::We simulate the inhibition of Ia-glutamatergic excitatory postsynaptic potential (EPSP) by preceding it with glycinergic recurrent (REN) and reciprocal (REC) inhibitory postsynaptic potentials (IPSPs). The inhibition is evaluated in the presence of voltage-dependent conductances of sodium, delayed rectifier potassium,...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00375

    authors: Gradwohl G,Grossman Y

    更新日期:2013-01-01 00:00:00

  • Dissociable forms of repetition priming: a computational model.

    abstract::Nondeclarative memory and novelty processing in the brain is an actively studied field of neuroscience, and reducing neural activity with repetition of a stimulus (repetition suppression) is a commonly observed phenomenon. Recent findings of an opposite trend-specifically, rising activity for unfamiliar stimuli-questi...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00569

    authors: Makukhin K,Bolland S

    更新日期:2014-04-01 00:00:00

  • Regularized neural networks: some convergence rate results.

    abstract::In a recent paper, Poggio and Girosi (1990) proposed a class of neural networks obtained from the theory of regularization. Regularized networks are capable of approximating arbitrarily well any continuous function on a compactum. In this paper we consider in detail the learning problem for the one-dimensional case. W...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.1995.7.6.1225

    authors: Corradi V,White H

    更新日期:1995-11-01 00:00:00

  • Abstract stimulus-specific adaptation models.

    abstract::Many neurons that initially respond to a stimulus stop responding if the stimulus is presented repeatedly but recover their response if a different stimulus is presented. This phenomenon is referred to as stimulus-specific adaptation (SSA). SSA has been investigated extensively using oddball experiments, which measure...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00077

    authors: Mill R,Coath M,Wennekers T,Denham SL

    更新日期:2011-02-01 00:00:00

  • Temporal coding: assembly formation through constructive interference.

    abstract::Temporal coding is studied for an oscillatory neural network model with synchronization and acceleration. The latter mechanism refers to increasing (decreasing) the phase velocity of each unit for stronger (weaker) or more coherent (decoherent) input from the other units. It has been demonstrated that acceleration gen...

    journal_title:Neural computation

    pub_type: 信件

    doi:10.1162/neco.2008.09-06-342

    authors: Burwick T

    更新日期:2008-07-01 00:00:00

  • On the classification capability of sign-constrained perceptrons.

    abstract::The perceptron (also referred to as McCulloch-Pitts neuron, or linear threshold gate) is commonly used as a simplified model for the discrimination and learning capability of a biological neuron. Criteria that tell us when a perceptron can implement (or learn to implement) all possible dichotomies over a given set of ...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.2008.20.1.288

    authors: Legenstein R,Maass W

    更新日期:2008-01-01 00:00:00

  • Conductance-based integrate-and-fire models.

    abstract::A conductance-based model of Na+ and K+ currents underlying action potential generation is introduced by simplifying the quantitative model of Hodgkin and Huxley (HH). If the time course of rate constants can be approximated by a pulse, HH equations can be solved analytically. Pulse-based (PB) models generate action p...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.1997.9.3.503

    authors: Destexhe A

    更新日期:1997-04-01 00:00:00

  • The dynamics of discrete-time computation, with application to recurrent neural networks and finite state machine extraction.

    abstract::Recurrent neural networks (RNNs) can learn to perform finite state computations. It is shown that an RNN performing a finite state computation must organize its state space to mimic the states in the minimal deterministic finite state machine that can perform that computation, and a precise description of the attracto...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.1996.8.6.1135

    authors: Casey M

    更新日期:1996-08-15 00:00:00

  • Does high firing irregularity enhance learning?

    abstract::In this note, we demonstrate that the high firing irregularity produced by the leaky integrate-and-fire neuron with the partial somatic reset mechanism, which has been shown to be the most likely candidate to reflect the mechanism used in the brain for reproducing the highly irregular cortical neuron firing at high ra...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00090

    authors: Christodoulou C,Cleanthous A

    更新日期:2011-03-01 00:00:00

  • Correlational Neural Networks.

    abstract::Common representation learning (CRL), wherein different descriptions (or views) of the data are embedded in a common subspace, has been receiving a lot of attention recently. Two popular paradigms here are canonical correlation analysis (CCA)-based approaches and autoencoder (AE)-based approaches. CCA-based approaches...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00801

    authors: Chandar S,Khapra MM,Larochelle H,Ravindran B

    更新日期:2016-02-01 00:00:00

  • Linking Neuromodulated Spike-Timing Dependent Plasticity with the Free-Energy Principle.

    abstract::The free-energy principle is a candidate unified theory for learning and memory in the brain that predicts that neurons, synapses, and neuromodulators work in a manner that minimizes free energy. However, electrophysiological data elucidating the neural and synaptic bases for this theory are lacking. Here, we propose ...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00862

    authors: Isomura T,Sakai K,Kotani K,Jimbo Y

    更新日期:2016-09-01 00:00:00

  • Methods for combining experts' probability assessments.

    abstract::This article reviews statistical techniques for combining multiple probability distributions. The framework is that of a decision maker who consults several experts regarding some events. The experts express their opinions in the form of probability distributions. The decision maker must aggregate the experts' distrib...

    journal_title:Neural computation

    pub_type: 杂志文章,评审

    doi:10.1162/neco.1995.7.5.867

    authors: Jacobs RA

    更新日期:1995-09-01 00:00:00

  • Representation sharpening can explain perceptual priming.

    abstract::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 ...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.2009.04-09-999

    authors: Moldakarimov S,Bazhenov M,Sejnowski TJ

    更新日期:2010-05-01 00:00:00

  • The computational structure of spike trains.

    abstract::Neurons perform computations, and convey the results of those computations through the statistical structure of their output spike trains. Here we present a practical method, grounded in the information-theoretic analysis of prediction, for inferring a minimal representation of that structure and for characterizing it...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/neco.2009.12-07-678

    authors: Haslinger R,Klinkner KL,Shalizi CR

    更新日期:2010-01-01 00:00:00

  • Regulation of ambient GABA levels by neuron-glia signaling for reliable perception of multisensory events.

    abstract::Activities of sensory-specific cortices are known to be suppressed when presented with a different sensory modality stimulus. This is referred to as cross-modal inhibition, for which the conventional synaptic mechanism is unlikely to work. Interestingly, the cross-modal inhibition could be eliminated when presented wi...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/NECO_a_00356

    authors: Hoshino O

    更新日期:2012-11-01 00:00:00

  • Synchronized firings in the networks of class 1 excitable neurons with excitatory and inhibitory connections and their dependences on the forms of interactions.

    abstract::Synchronized firings in the networks of class 1 excitable neurons with excitatory and inhibitory connections are investigated, and their dependences on the forms of interactions are analyzed. As the forms of interactions, we treat the double exponential coupling and the interactions derived from it: pulse coupling, ex...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/0899766053630387

    authors: Kanamaru T,Sekine M

    更新日期:2005-06-01 00:00:00

  • A neural-network-based approach to the double traveling salesman problem.

    abstract::The double traveling salesman problem is a variation of the basic traveling salesman problem where targets can be reached by two salespersons operating in parallel. The real problem addressed by this work concerns the optimization of the harvest sequence for the two independent arms of a fruit-harvesting robot. This a...

    journal_title:Neural computation

    pub_type: 杂志文章

    doi:10.1162/08997660252741194

    authors: Plebe A,Anile AM

    更新日期:2002-02-01 00:00:00