The book of Hebb

Communication in neuronal networks

Brains perform with remarkable efficiency, are capable of prodigious computation, and are marvels of communication. We are beginning to understand some of the geometric, biophysical, and energy constraints that have governed the evolution of cortical networks. To operate efficiently within these constraints, nature has optimized the structure and function of cortical networks with design principles similar to those used in electronic networks. The brain also exploits the adaptability of biological systems to reconfigure in response to changing needs.

Self-organizing neural systems based on predictive learning

The ability to predict future events based on the past is an important attribute of organisms that engage in adaptive behaviour. One prominent computational method for learning to predict is called temporal-difference (TD) learning. It is so named because it uses the difference between successive predictions to learn to predict correctly. TD learning is well suited to modelling the biological phenomenon of conditioning, wherein an organism learns to predict a reward even though the reward may occur later in time. We review a model for conditioning in bees based on TD learning. The model illustrates how the TD-learning algorithm allows an organism to learn an appropriate sequence of actions leading up to a reward, based solely on reinforcement signals. The second part of the paper describes how TD learning can be used at the cellular level to model the recently discovered phenomenon of spike-timing-dependent plasticity. Using a biophysical model of a neocortical neuron, we demonstrate that the shape of the spike-timing-dependent learning windows found in biology can be interpreted as a form of TD learning occurring at the cellular level. We conclude by showing that such spike-based TD-learning mechanisms can produce direction selectivity in visual-motion-sensitive cells and can endow recurrent neocortical circuits with the powerful ability to predict their inputs at the millisecond time-scale.

Neural mechanisms for the recognition of biological movements

The visual recognition of complex movements and actions is crucial for the survival of many species. It is important not only for communication and recognition at a distance, but also for the learning of complex motor actions by imitation. Movement recognition has been studied in psychophysical, neurophysiological and imaging experiments, and several cortical areas involved in it have been identified. We use a neurophysiologically plausible and quantitative model as a tool for organizing and making sense of the experimental data, despite their growing size and complexity. We review the main experimental findings and discuss possible neural mechanisms, and show that a learning-based, feedforward model provides a neurophysiologically plausible and consistent summary of many key experimental results.

A problem with Hebb and local spikes

Although our understanding of the cellular properties of mammalian neurons is increasing rapidly, the computational function of their elaborate dendritic trees is still mysterious. In recent years, experiments have shown that, in pyramidal cells, individual dendritic compartments sustain local excitation spikes.. These dendrites also support Hebbian synaptic plasticity, which depends on the precise temporal relationship between pre- and postsynaptic spikes. In this review we discuss what we consider to be a problem with Hebbian (i.e., spike-timing-dependent) plasticity. We argue that most of the spikes that occur in dendrites are not back-propagating action potentials but but local spikes, and that Hebbian plasticity caused by local spikes can undermine the functional integrity of the geometrically complex dendritic tree. We propose that the inverted Hebbian plasticity of synapses involved in local spikes, and/or local dendritic homeostatic plasticity, could prevent an unbalanced distribution of synaptic weights on the dendritic tree.

Optical imaging of long-lasting depolarization on burst stimulation in area CA1 of rat hippocampal slices

Postsynaptic depolarization of dendrites paired with spike generation at the soma is considered to be a central mechanism of long-term potentiation (LTP) induction and a prime example of a Hebbian synapse. This pairing, however, has never been actually demonstrated on tetanic stimulation. Optical imaging of neural activity with a voltage-sensitive dye (VSD) is one potentially suitable method for examining this pairing. It is possible with optical recording to examine simultaneously the excitation of postsynaptic neurons at multiple sites. Thus the pairing of spike generation at the soma and dendritic depolarization can be examined with population level optical recording in highly laminar structures such as the hippocampal slice preparation. For example, one can correlate the optical signals obtained from cell layers with the activity of the soma, and, similarly, optical signals from stratum radiatum can be correlated with the activity of the apical dendrite, even though one cannot calibrate the optical signals in terms of actual membrane potential. Using the VSD aminonaphthylethenylpyridinium in rat hippocampal slices, we aimed to examine the pairing. Standard tetanic stimulation (100 Hz, 1 s) that elicited LTP in the field excitatory postsynaptic potential (fEPSP) resulted in a long-lasting depolarizing optical signal (about 2 s) that spread progressively along the known input pathway of CA1. The time course of this long-lasting depolarization was similar to that recorded intracellularly and to that reflected in the fEPSP. The long-lasting depolarization was insensitive to D,L-2-amino-5-phosphonovaleric acid (D,L-APV, 50 microM), but D,L-APV inhibited the induction of LTP; this allowed us to increase the signal-to-noise ratio of the optical signal by averaging several trials. Using this improved optical signal, we confirmed that postsynaptic cells practically "missed" spikes during tetanic stimulation in most parts of CA1, which had been suggested in the intracellular recordings. Intracellular recordings revealed a 23% reduction in input resistance, which might explain the failed spike generation at the soma via shunting. A steep spatial convergence of the depolarization along the transverse axis of area CA1 was observed. In contrast to the response resulting from a standard 100-Hz tetanus, broader activation, and paired depolarization with somatic spikes was observed on theta-burst stimulation. Overall we concluded that postsynaptic spike generation, at least in synchronous form, has less effect on LTP induction with standard tetanic stimulation, while theta-burst tetanic stimulation can elicit pairing of dendritic depolarization and somatic discharge.

Mechanisms for temporal tuning and filtering by postsynaptic signaling pathways

Networks of signaling pathways perform complex temporal decoding functions in diverse biological systems, including the synapse, development, and bacterial chemotaxis. This paper examines temporal filtering and tuning properties of synaptic signaling pathways as a possible substrate for emergent temporal decoding. A mass action kinetic model of 16 synaptic signaling pathways was used to dissect out the contribution of these pathways in linear cascades and when coupled to form a network. The model predicts two primary mechanisms of temporal tuning of pathways: a weighted summation of responses of pathways with different timings and the presence of biochemical feedback loop(s) with emergent dynamics. Regulatory inputs act differently on these two tuning mechanisms. In the first case, regulators act like a gain-control on pathways with different intrinsic tuning. In the case of feedback loops, the temporal properties of the loop itself are changed. These basic tuning mechanisms may underlie specialized temporal tuning functions in more complex signaling systems in biology.

Matters temporal

Current evidence suggests that neural Hebbian learning in cortical and hippocampal synapses is fundamentally predictive rather than conventionally correlational. Much attention is focussing on what sort of predictions are acquired, and in what neural architectures. A recent paper by Rao and Sejnowski has suggested an interesting interpretation in terms of a popular predictive algorithm that has roots in psychology, computer science and engineering.

Synaptic modification in neural circuits: a timely action

Long-term modification of synaptic strength is thought to be the basic mechanism underlying the activity-dependent refinement of neural circuits and the formation of memories engrammed on them. Studies ranging from cell culture preparations to humans subjects indicate that the decision of whether a synapse will undergo strengthening or weakening critically depends on the temporal order of presynaptic and postsynaptic activity. At many synapses, potentiation will be induced only when the presynaptic neuron fires an action potential within milliseconds before the postsynaptic neuron fires, whereas weakening will occur when it is the postsynaptic neuron that fires first. Such processes might be important for the remodeling of neural circuits by activity during development and for network functions such as sequence learning and prediction. Ultimately, this synaptic property might also be fundamental for the cognitive process by which we structure our experience through cause and effect relations.

The effect of gaze angle and fixation distance on the responses of neurons in V1, V2, and V4

What we see depends on where we look. This paper characterizes the modulatory effects of point of regard in three-dimensional space on responsiveness of visual cortical neurons in areas V1, V2, and V4. Such modulatory effects are both common, affecting 85% of cells, and strong, frequently producing changes of mean firing rate by a factor of 10. The prevalence of neurons in area V4 showing a preference for near distances may be indicative of the involvement of this area in close scrutiny during object recognition. We propose that eye-position signals can be exploited by visual cortex as classical conditioning stimuli, enabling the perceptual learning of systematic relationships between point of regard and the structure of the visual environment.

Spike-timing-dependent Hebbian plasticity as temporal difference learning

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.

Synaptic modification by correlated activity: Hebb's postulate revisited

Correlated spiking of pre- and postsynaptic neurons can result in strengthening or weakening of synapses, depending on the temporal order of spiking. Recent findings indicate that there are narrow and cell type-specific temporal windows for such synaptic modification and that the generally accepted input- (or synapse-) specific rule for modification appears not to be strictly adhered to. Spike timing-dependent modifications, together with selective spread of synaptic changes, provide a set of cellular mechanisms that are likely to be important for the development and functioning of neural networks. When an axon of cell A is near enough to excite cell B or repeatedly or consistently takes part in firing it, some growth or metabolic change takes place in one or both cells such that A's efficiency, as one of the cells firing B, is increased.

Associative long-term depression in the hippocampus is dependent on postsynaptic N-type Ca2+ channels

Long-term depression (LTD) is a form of synaptic plasticity that can be induced either by low-frequency stimulation of presynaptic fibers or in an associative manner by asynchronous pairing of presynaptic and postsynaptic activity. We investigated the induction mechanisms of associative LTD in CA1 pyramidal neurons of the hippocampus using whole-cell patch-clamp recordings and Ca(2+) imaging in acute brain slices. Asynchronous pairing of postsynaptic action potentials with EPSPs evoked with a delay of 20 msec induced a robust, long-lasting depression of the EPSP amplitude to 43%. Unlike LTD induced by low-frequency stimulation, associative LTD was resistant to the application of d-AP-5, indicating that it is independent of NMDA receptors. In contrast, associative LTD was inhibited by (S)-alpha-methyl-4-carboxyphenyl-glycine, indicating the involvement of metabotropic glutamate receptors. Furthermore, associative LTD is dependent on the activation of voltage-gated Ca(2+) channels by postsynaptic action potentials. Both nifedipine, an L-type Ca(2+) channel antagonist, and omega-conotoxin GVIA, a selective N-type channel blocker, abolished the induction of associative LTD. 8-hydroxy-2-dipropylaminotetralin (OH-DPAT), a 5-HT(1A) receptor agonist, inhibited postsynaptic Ca(2+) influx through N-type Ca(2+) channels, without affecting presynaptic transmitter release. OH-DPAT also inhibited the induction of associative LTD, suggesting that the involvement of N-type channels makes synaptic plasticity accessible to modulation by neurotransmitters. Thus, the modulation of N-type Ca(2+) channels provides a gain control for synaptic depression in hippocampal pyramidal neurons.

Disparity in neurotransmitter release probability among competing inputs during neuromuscular synapse elimination

Competition among the several motor axons transiently innervating neonatal muscle fibers results in an increasing disparity in the quantal content and synaptic territory of each competitor, culminating in the permanent loss of all but one axon from neuromuscular junctions. We asked whether differences in the probability of neurotransmitter release also contribute to the increasing disparity in quantal content among competing inputs, and when in the process of competition changes in release probability become apparent. To address these questions, intracellular recordings were made from dually innervated neonatal mouse soleus muscle fibers, and quantal content and paired-pulse facilitation were evaluated for each input. At short interpulse intervals, paired-pulse facilitation was significantly higher for the weaker input with the smaller quantal content than the stronger input with the larger quantal content. Because neurotransmitter release probability across all release sites is inversely related to the extent of facilitation observed after paired-pulse stimulation, this result suggests that release probability is lower for weak compared with strong inputs innervating the same junction. A disparity in the probability of neurotransmitter release thus contributes to the disparity in quantal content that occurs during synaptic competition. Together, this work suggests that an input incapable of sustaining a high release probability may be at a competitive disadvantage for synaptic maintenance.

Impact of correlated synaptic input on output firing rate and variability in simple neuronal models

Cortical neurons are typically driven by thousands of synaptic inputs. The arrival of a spike from one input may or may not be correlated with the arrival of other spikes from different inputs. How does this interdependence alter the probability that the postsynaptic neuron will fire? We constructed a simple random walk model in which the membrane potential of a target neuron fluctuates stochastically, driven by excitatory and inhibitory spikes arriving at random times. An analytic expression was derived for the mean output firing rate as a function of the firing rates and pairwise correlations of the inputs. This stochastic model made three quantitative predictions. (1) Correlations between pairs of excitatory or inhibitory inputs increase the fluctuations in synaptic drive, whereas correlations between excitatory-inhibitory pairs decrease them. (2) When excitation and inhibition are fully balanced (the mean net synaptic drive is zero), firing is caused by the fluctuations only. (3) In the balanced case, firing is irregular. These theoretical predictions were in excellent agreement with simulations of an integrate-and-fire neuron that included multiple conductances and received hundreds of synaptic inputs. The results show that, in the balanced regime, weak correlations caused by signals shared among inputs may have a multiplicative effect on the input-output rate curve of a postsynaptic neuron, i.e. they may regulate its gain; in the unbalanced regime, correlations may increase firing probability mainly around threshold, when output rate is low; and in all cases correlations are expected to increase the variability of the output spike train.

Timing-based LTP and LTD at vertical inputs to layer II/III pyramidal cells in rat barrel cortex

Experience-dependent plasticity in somatosensory (S1) and visual (V1) cortex involves rapid depression of responses to a deprived sensory input (a closed eye or a trimmed whisker). Such depression occurs first in layer II/III and may reflect plasticity at vertical inputs from layer IV to layer II/III pyramids. Here, I describe a timing-based, associative form of long-term potentiation and depression (LTP/LTD) at this synapse in S1. LTP occurred when excitatory postsynaptic potentials (EPSPs) led single postsynaptic action potentials (APs) within a narrow temporal window, and LTD occurred when APs led EPSPs within a significantly broader window. This long LTD window is unusual among timing-based learning rules and causes EPSPs that are uncorrelated with postsynaptic APs to become depressed. This behavior suggests a simple model for depression of deprived sensory responses in S1 and V1.

Natural patterns of activity and long-term synaptic plasticity

Long-term potentiation (LTP) of synaptic transmission is traditionally elicited by massively synchronous, high-frequency inputs, which rarely occur naturally. Recent in vitro experiments have revealed that both LTP and long-term depression (LTD) can arise by appropriately pairing weak synaptic inputs with action potentials in the postsynaptic cell. This discovery has generated new insights into the conditions under which synaptic modification may occur in pyramidal neurons in vivo. First, it has been shown that the temporal order of the synaptic input and the postsynaptic spike within a narrow temporal window determines whether LTP or LTD is elicited, according to a temporally asymmetric Hebbian learning rule. Second, backpropagating action potentials are able to serve as a global signal for synaptic plasticity in a neuron compared with local associative interactions between synaptic inputs on dendrites. Third, a specific temporal pattern of activity--postsynaptic bursting--accompanies synaptic potentiation in adults.