The legacy of Donald O. Hebb: more than the Hebb synapse

Neuroscientists associate the name of Donald O. Hebb with the Hebbian synapse and the Hebbian learning rule, which underlie connectionist theories and synaptic plasticity, but Hebb's work has also influenced developmental psychology, neuropsychology, perception and the study of emotions, as well as learning and memory. Here, we review the work of Hebb and its lasting influence on neuroscience in honour of the 2004 centenary of his birth.     

Visual channels, Hebbian assemblies and the effect of Hebb's rule

 A model of sensory learning is proposed that is based upon Hebb's rule, where Hebb's rule has been generalised by introducing a stabilising function representing some feedback process within or at the adapting (cortical) neuron, preventing synaptic weights from increasing without limit. It will be shown that neurons adapting according to this stabilised Hebb rule will turn into a matched filter for that part of the stimulus pattern that covers the receptive field of a neuron. It follows that the presentation of a stimulus pattern may imply the formation of a set of neurons with overlapping receptive fields, where each neuron has adapted to a certain part of the stimulus. Making simplifying assumptions about the detection process, the model will be illustrated, fitting it to data from Meinhardt and Mortensen [Meinhardt G, Mortensen U (1998) Biol Cybern 79:413–425] which are not compatible with the classical matched filter model introduced by Hauske et al. [Hauske G, Wolf W, Lupp U (1976) Biol Cybern 22:181–188]. Received: 10 May 1999 / Accepted in revised form: 22 October 1999     

Hebbian learning reconsidered: Representation of static and dynamic objects in associative neural nets

According to Hebb's postulate for learning, information presented to a neural net during a learning session is stored in the synaptic efficacies. Long-term potentiation occurs only if the postsynaptic neuron becomes active in a time window set up by the presynaptic one. We carefully interpret and mathematically implement the Hebb rule so as to handle both stationary and dynamic objects such as single patterns and cycles. Since the natural dynamics contains a rather broad distribution of delays, the key idea is to incorporate these delays in the learning session. As theory and numerical simulation show, the resulting procedure is surprisingly robust and faithful. It also turns out that pure Hebbian learning is by selection: the network produces synaptic representations that are selected according to their resonance with the input percepts.     

近存储饱和状态下联想学习记忆的神经网络模型

本文提出了神经网络在近饱和状态下的一种联想学习记忆模型.讨论了该模型的主要特性,对由100个神经元、记忆10个随机图样组成的网络系统给出并分析了计算机模拟结果,讨论了该模型的学习律与传统的Hebb学习律的区别,研究了网络在学习记忆和联想新态时初始噪声Pi和联想噪声Pa对新态恢复行为的影响,总结了在近饱和状态下该模型所具有的优势.     

The spatiotemporal learning rule and its efficiency in separating spatiotemporal patterns

The hippocampus plays an important role in the course of establishing long-term memory, i.e., to make short-term memory of spatially and temporally associated input information. In 1996 (Tsukada et al. 1996), the spatiotemporal learning rule was proposed based on differences observed in hippocampal long-term potentiation (LTP) induced by various spatiotemporal pattern stimuli. One essential point of this learning rule is that the change of synaptic weight depends on both spatial coincidence and the temporal summation of input pulses. We applied this rule to a single-layered neural network and compared its ability to separate spatiotemporal patterns with that of other rules, including the Hebbian learning rule and its extended rules. The simulated results showed that the spatiotemporal learning rule had the highest efficiency in discriminating spatiotemporal pattern sequences, while the Hebbian learning rule (including its extended rules) was sensitive to differences in spatial patterns.     

Input-dependent learning rule for the memory of spatiotemporal sequences in hippocampal network with theta phase precession

Theta phase precession in rat hippocampal place cells is hypothesized to contribute to memory encoding of running experience in the sense that it provides the ideal timing for synaptic plasticity and enables the asymmetric associative connections under the Hebbian learning rule with asymmetric time window (Yamaguchi 2003). When the sequence of place fields is considered as the episodic memory of running experience, a given spatial route should be accurately stored in spite of differing overlap extent among place fields and varying running velocity. Using a hippocampal network model with phase precession and the Hebbian learning rule with asymmetric time window, we investigate the memory encoding of place field sequences in a single traversal experience. Computer experiments show that place fields cannot be stored correctly until an input-dependent feature is introduced into the learning rule. These experiments further indicate that there exists an optimum value for the saturation level of synaptic plasticity and the speed of synaptic plasticity in the learning rule, which are correlated with, respectively, the overlap extent of place field sequence and the running velocity of animal during traversal. A comparison of these results with biological evidences shows good agreement and suggests that behavior-dependent regulation of the learning rule is necessary for memory encoding.     

Optimal learning rules for familiarity detection

It has been suggested that the mammalian memory system has both familiarity and recollection components. Recently, a high-capacity network to store familiarity has been proposed. Here we derive analytically the optimal learning rule for such a familiarity memory using a signal- to-noise ratio analysis. We find that in the limit of large networks the covariance rule, known to be the optimal local, linear learning rule for pattern association, is also the optimal learning rule for familiarity discrimination. In the limit of large networks, the capacity is independent of the sparseness of the patterns and the corresponding information capacity is 0.057 bits per synapse, which is somewhat less than typically found for associative networks.     

Learning the evolutionarily stable strategy

The possibility that animals learn a “developmentally stable strategy” (DSS) (Dawkins, 1980) is an alternative in biological game theory to the idea that evolutionarily stable strategies (ESS) (Maynard Smith, 1972) are genetically determined. A learning rule is defined as a rule which assigns for every possible behaviour the probability of displaying that behaviour at each trial of a game as a function of previous payoffs. This report examines properties of the evolutionarily stable (ES) learning rule, i.e. the rule which, when adopted by a population, is uninvadable by a mutant with a different learning rule. The DSS is defined as the strategy used by individuals with the ES learning rule. With some simplifying assumptions, it is shown that the DSS is the ESS: the ES learning rule is a rule for learning ESSs. This and other properties of the ES learning rule suggested that an approximation to such a rule is the relative payoff sum (RPS) learning rule, which states that the probability of displaying a behaviour is equal to the cumulative payoff for that behaviour relative to the total sum of payoffs for the game. Residual payoffs and a memory factor are incorporated into the RPS learning rule to account for prior expectations of payoff and the decay of memory with time. Both features are adaptive. In simulations of several frequency dependent and frequency independent games using the RPS learning rule, the response of the simulated animals was consistent with the predictions of the ES learning rule. This analysis has shown how ESSs may be achieved by non-genetic means. The RPS learning rule is described in molecular terms utilizing synthesis, storage, and degradation of a substance which elicits the behavioural response. If the RPS learning rule is used by animals, it should be possible to identify within neurons substances whose synthesis is regulated by behavioural stimuli and which initiate alternative behaviours in proportion to their concentrations.     

Neural networks through the hourglass

The effect of the synaptic plasticity on the dynamics of a large neural network is studied. Our approach is analytic but inspired by the data, both simulated and experimental. We explain formation of the small strongly connected assemblies within a dynamical network following Hebb's rule. Also, we find the conditions for the synchrony effect in the stochastic network in the absence of large synchronized input.     

Hippocampal synaptic plasticity: role in spatial learning or the automatic recording of attended experience?

Allocentric spatial learning can sometimes occur in one trial. The incorporation of information into a spatial representation may, therefore, obey a one-trial correlational learning rule rather than a multi-trial error-correcting rule. It has been suggested that physiological implementation of such a rule could be mediated by N-methyl-D-aspartate (NMDA) receptor-dependent long-term potentiation (LTP) in the hippocampus, as its induction obeys a correlational type of synaptic learning rule. Support for this idea came originally from the finding that intracerebral infusion of the NMDA antagonist AP5 impairs spatial learning, but studies summarized in the first part of this paper have called it into question. First, rats previously given experience of spatial learning in a watermaze can learn a new spatial reference memory task at a normal rate despite an appreciable NMDA receptor blockade. Second, the classical phenomenon of ''blocking'' occurs in spatial learning. The latter finding implies that spatial learning can also be sensitive to an animal''s expectations about reward and so depend on more than the detection of simple spatial correlations. In this paper a new hypothesis is proposed about the function of hippocampal LTP. This hypothesis retains the idea that LTP subserves rapid one-trial memory, but abandons the notion that it serves any specific role in the geometric aspects of spatial learning. It is suggested that LTP participates in the automatic recording of attended experience'': a subsystem of episodic memory in which events are temporarily remembered in association with the contexts in which they occur. An automatic correlational form of synaptic plasticity is ideally suited to the online registration of context event associations. In support, it is reported that the ability of rats to remember the most recent place they have visited in a familiar environment is exquisitely sensitive to AP5 in a delay-dependent manner. Moreover, new studies of the lasting persistence of NMDA-dependent LTP, known to require protein synthesis, point to intracellular mechanisms that enable transient synaptic changes to be stabilized if they occur in close temporal proximity to important events. This new property of hippocampal LTP is a desirable characteristic of an event memory system.     

Spatiotemporal specificity of synaptic plasticity: cellular rules and mechanisms

Recent experimental results on spike-timing-dependent plasticity (STDP) and heterosynaptic interaction in various systems have revealed new temporal and spatial properties of activity-dependent synaptic plasticity. These results challenge the conventional understanding of Hebb's rule and raise intriguing questions regarding the fundamental processes of cellular signaling. In this article, I review these new findings that lead to formulation of a new set of cellular rules. Emphasis is on evaluating potential molecular and cellular mechanisms that may underlie the spike-timing window of STDP and different patterns of heterosynaptic modifications. I also highlight several unresolved issues, and suggest future lines of research.     

Activity-dependent wiring of the developing hippocampal neuronal circuit

In the developing hippocampus, functional excitatory synaptic connections seem to be recruited from a preformed, initially silent synaptic network. This functional synapse induction requires presynaptic action potentials paired with postsynaptic depolarization, thus obeying Hebb's rule of association. During early postnatal development the hippocampus exhibits an endogenous form of patterned neuronal activity that is driven by GABAergic depolarization. We propose that this recurrent activity promotes the input-specific induction of functional synapses in the CA1 region. Thus, activity-dependent synaptic reorganization in the developing hippocampus appears to be dominated by an active recruitment of new synapses rather than an active elimination of redundant connections.     

Precise-Spike-Driven Synaptic Plasticity: Learning Hetero-Association of Spatiotemporal Spike Patterns

A new learning rule (Precise-Spike-Driven (PSD) Synaptic Plasticity) is proposed for processing and memorizing spatiotemporal patterns. PSD is a supervised learning rule that is analytically derived from the traditional Widrow-Hoff rule and can be used to train neurons to associate an input spatiotemporal spike pattern with a desired spike train. Synaptic adaptation is driven by the error between the desired and the actual output spikes, with positive errors causing long-term potentiation and negative errors causing long-term depression. The amount of modification is proportional to an eligibility trace that is triggered by afferent spikes. The PSD rule is both computationally efficient and biologically plausible. The properties of this learning rule are investigated extensively through experimental simulations, including its learning performance, its generality to different neuron models, its robustness against noisy conditions, its memory capacity, and the effects of its learning parameters. Experimental results show that the PSD rule is capable of spatiotemporal pattern classification, and can even outperform a well studied benchmark algorithm with the proposed relative confidence criterion. The PSD rule is further validated on a practical example of an optical character recognition problem. The results again show that it can achieve a good recognition performance with a proper encoding. Finally, a detailed discussion is provided about the PSD rule and several related algorithms including tempotron, SPAN, Chronotron and ReSuMe.     

Long term memory storage capacity of multiconnected neural networks

Quantitative expressions of long-term memory storage capacities of complex neural network are derived. The networks are made of neurons connected by synapses of any order, of the axono-axonal type considered by Kandel et al. for example. The effect of link deletion possibly related to aging, is also considered. The central result of this study is that, within the framework of Hebb's laws, the number of stored bits is proportional to the number of synapses. The proportionality factor however, decreases when the order of involved synaptic contact increases. This tends to favor neural architectures with low-order synaptic connectivities. It is finally shown that the memory storage capacities can be optimized by a partition of the network into neuron clusters with size comparable with that observed for cortical microcolumns.     

Imprinting as a mechanism of information memorizing in the adult BALB/c mice

Study of spatial learning in adult BALB/c mice revealed that a short exposition to the environment (from 3 to 8 minutes) could be enough for spatial information to be fixed in the long-term memory, and affected subsequent learning process in the new environment. Control group, learning in the same maze, followed the "shortest path" principle during formation of the optimal food-obtaining habit. Experimental animals, learning in a slightly changed environment, were unable to apply this rule due to persistent coupling of the new spatial information with the old memory traces which led to constant errors. The obtained effect was observed during the whole learning period and depended neither on frequency nor on interval of repetition during the initial information acquisition. The obtained data testify that memorizing in adult state share the properties with the imprinting process inherent in the early ontogeny. The memory fixation on all development stages seems to be based on a universal mechanism.     

Learning of oscillatory correlated patterns in a cortical network by a STDP-based learning rule

In this paper, we propose an iterative learning rule that allows the imprinting of correlated oscillatory patterns in a model of the hippocampus able to work as an associative memory for oscillatory spatio-temporal patterns. We analyze the dynamics in the Fourier domain, showing how the network selectively amplify or distort the Fourier components of the input, in a manner which depends on the imprinted patterns. We also prove that the proposed iterative local rule converges to the pseudo-inverse rule generalized to oscillatory patterns.     

LTD windows of the STDP learning rule and synaptic connections having a large transmission delay enable robust sequence learning amid background noise

Spike-timing-dependent synaptic plasticity (STDP) is a simple and effective learning rule for sequence learning. However, synapses being subject to STDP rules are readily influenced in noisy circumstances because synaptic conductances are modified by pre- and postsynaptic spikes elicited within a few tens of milliseconds, regardless of whether those spikes convey information or not. Noisy firing existing everywhere in the brain may induce irrelevant enhancement of synaptic connections through STDP rules and would result in uncertain memory encoding and obscure memory patterns. We will here show that the LTD windows of the STDP rules enable robust sequence learning amid background noise in cooperation with a large signal transmission delay between neurons and a theta rhythm, using a network model of the entorhinal cortex layer II with entorhinal-hippocampal loop connections. The important element of the present model for robust sequence learning amid background noise is the symmetric STDP rule having LTD windows on both sides of the LTP window, in addition to the loop connections having a large signal transmission delay and the theta rhythm pacing activities of stellate cells. Above all, the LTD window in the range of positive spike-timing is important to prevent influences of noise with the progress of sequence learning.     

How Attention Can Create Synaptic Tags for the Learning of Working Memories in Sequential Tasks

Intelligence is our ability to learn appropriate responses to new stimuli and situations. Neurons in association cortex are thought to be essential for this ability. During learning these neurons become tuned to relevant features and start to represent them with persistent activity during memory delays. This learning process is not well understood. Here we develop a biologically plausible learning scheme that explains how trial-and-error learning induces neuronal selectivity and working memory representations for task-relevant information. We propose that the response selection stage sends attentional feedback signals to earlier processing levels, forming synaptic tags at those connections responsible for the stimulus-response mapping. Globally released neuromodulators then interact with tagged synapses to determine their plasticity. The resulting learning rule endows neural networks with the capacity to create new working memory representations of task relevant information as persistent activity. It is remarkably generic: it explains how association neurons learn to store task-relevant information for linear as well as non-linear stimulus-response mappings, how they become tuned to category boundaries or analog variables, depending on the task demands, and how they learn to integrate probabilistic evidence for perceptual decisions.     

Behavioral characterization of mice lacking the ubiquitin ligase UBR1 of the N-end rule pathway

The N-end rule pathway, a subset of the ubiquitin pathway, relates the in vivo half-life of a protein to the identity of its N-terminal residue. Mice lacking NTAN1, a component of the N-end rule pathway, showed altered learning and memory, and socially conditioned behavioral alteration ( Balogh, Kwon, & Denenberg 1999, 2000 ; Kwon, Balogh et al . 2000 ; Balogh et al . 2001 ). Mice lacking UBR1 (E3α), one of at least three recognition components of the N-end rule pathway, are viable and outwardly normal. Here we describe behavioral characterizations of UBR1 knockout (UBR1^–/–) mice. Compared to congenic littermates, UBR1^–/– mice exhibited less spontaneous activity in an open field and took longer to locate the hidden platform during eight-week Morris water maze retention. In contrast, they performed better in horizontal–vertical discrimination and Lashley III maze testing. No statistically significant differences in inhibitory learning were observed. With the exception of enhanced Lashley III maze performance, these data parallel findings with NTAN1^–/– mice lacking the upstream component of UBR1. These results suggest that, like NTAN1, UBR1 is involved in learning and memory.     

Rule-Based Category Learning in Children: The Role of Age and Executive Functioning

Rule-based category learning was examined in 4–11 year-olds and adults. Participants were asked to learn a set of novel perceptual categories in a classification learning task. Categorization performance improved with age, with younger children showing the strongest rule-based deficit relative to older children and adults. Model-based analyses provided insight regarding the type of strategy being used to solve the categorization task, demonstrating that the use of the task appropriate strategy increased with age. When children and adults who identified the correct categorization rule were compared, the performance deficit was no longer evident. Executive functions were also measured. While both working memory and inhibitory control were related to rule-based categorization and improved with age, working memory specifically was found to marginally mediate the age-related improvements in categorization. When analyses focused only on the sample of children, results showed that working memory ability and inhibitory control were associated with categorization performance and strategy use. The current findings track changes in categorization performance across childhood, demonstrating at which points performance begins to mature and resemble that of adults. Additionally, findings highlight the potential role that working memory and inhibitory control may play in rule-based category learning.