Rolls & Deco; Noisy Brain
Book Page   Topic    
Rolls & Deco; Noisy Brain 0 The relatively random spiking times of individual neurons produces a source of noise in the brain.
Rolls & Deco; Noisy Brain 0 Functional neuroimaging is useful to indicate where in the human brain different processes take place, and the show which functions can be disassociated from each other. 0
Rolls & Deco; Noisy Brain 3 There are often 5000-20,000 inputs per neuron. 3
Rolls & Deco; Noisy Brain 4 A learning rule presaged by Donald Hebb (1949) proposes that synapses increase in strength when there is conjunctive presynaptic and postsynaptic activity. 1
Rolls & Deco; Noisy Brain 5 Long-term potentiation is a use-dependent and sustained increase in synaptic strength that can be induced by a brief periods of synaptic stimulation. 1
Rolls & Deco; Noisy Brain 8 LTP is long-lasting, develops rapidly, is synapse specific, and in some cases is associative. It is of interest as a potential synaptic mechanisms underlying some forms of memory. 3
Rolls & Deco; Noisy Brain 12 Systems-level on Analysis of Brain Function 4
Rolls & Deco; Noisy Brain 13 Pathways involved in some sensory systems (diagram) 1
Rolls & Deco; Noisy Brain 17 Find structure and connectivity of the neocortex 4
Rolls & Deco; Noisy Brain 17 Excitatory cells in connections 0
Rolls & Deco; Noisy Brain 19 Inhibitory cells and connections 2
Rolls & Deco; Noisy Brain 19 Basket cells comprise 5-7% of the total cortical cell population, compared with approximately 72% of pyramidal cells. 0
Rolls & Deco; Noisy Brain 19 Basket cells receive synapses from the main extrinsic afferents to the neocortex, including thalamic afferents, so that they must contribute to a feedforward inhibition of pyramidal cells. 0
Rolls & Deco; Noisy Brain 22 Typical quantitative estimates for neocortex (table) 3
Rolls & Deco; Noisy Brain 22 Local recurrent connectivity implementing local reentrantcy is a feature of cortical computation. 0
Rolls & Deco; Noisy Brain 22 The number of recurrent connections per neuron is in the order of 10,000. 0
Rolls & Deco; Noisy Brain 23 A principle of cortical design is that it does have in part local connectivity, so that each part can have its own processing and storage, which may be triggered by other modules, but is a distinct operation from that which occurs simultaneously in other modules. 1
Rolls & Deco; Noisy Brain 23 Functional pathways through cortical layers. 0
Rolls & Deco; Noisy Brain 23 In primary sensory cortical areas, the main extrinsic forward input is from the thalamus, and ends in layer 4, where synapses are formed onto spiny stellate cells. These in turn project heavily onto pyramidal cells in layers 3 and 2, which in turn send projections forward to the next cortical area. 0
Rolls & Deco; Noisy Brain 23 The circuitry is more complex because the thalamic afferents also synapse onto the basal dendrites in or close to the layer 2 pyramidal cells, as well is onto layer 6 pyramidal cells and inhibitory interneurons. 0
Rolls & Deco; Noisy Brain 25 Forward corto-cortical afferents to a cortical area sometimes have a columnar pattern to their distribution, with a column width 200-300 in diameter. 2
Rolls & Deco; Noisy Brain 26 Theoretical Significance of Back Projections in the Neocortex 1
Rolls & Deco; Noisy Brain 26 Cortical processing takes place through a hierarchy of cortical stages. Convergence and competition are key aspects of the processing. 0
Rolls & Deco; Noisy Brain 26 Neurons in each cortical stage respond for 20-30 ms when an object can just be seen. 0
Rolls & Deco; Noisy Brain 26 The time for processing to travel from V1 to the inferior temporal visual cortex (IT) is approximately 50 ms. 0
Rolls & Deco; Noisy Brain 26 Backprojections are a major feature of cortical connectivity. 0
Rolls & Deco; Noisy Brain 26 Some of the functions of back projections are in memory recall, in forming attractor networks in which there are associatively modifiable synaptic connections in both the feedforward and feedback connections between adjacent levels in the hierarchy, and in top-down attentional modulation implemented by biased competition. 0
Rolls & Deco; Noisy Brain 27 Orbitofrontal cortex, amygdala and hippocampus are stages of information processing at which the different sensory modalities (such as vision, hearing, touch, taste, and smell for the orbitofrontal cortex and amygdala) are brought together, so that correlations between inputs in different modalities can be detected in these regions, but not at prior cortical processing stages in each modality, as these cortical processing stages are mainly unimodal. 1
Rolls & Deco; Noisy Brain 27 As a result of bringing together any two sensory modalities, significant correspondences between the two modalities can be detected. 0
Rolls & Deco; Noisy Brain 30 A highly developed system of excitatory recurrent collateral connections between nearby pyramidal cells is a hallmark of neocortical design. 3
Rolls & Deco; Noisy Brain 31 Autoassociative memories, or attractor neural networks, store memories, each one of which is represented by a pattern of neural activity. 1
Rolls & Deco; Noisy Brain 31 Memories are stored in the recurrent synaptic connections between the neurons of the network, for example in the recurrent collateral connections between cortical pyramidal cells. 0
Rolls & Deco; Noisy Brain 31 Autoassociative networks can recall the appropriate memory from the network when provided with a fragment of one of the memories. This is called completion. 0
Rolls & Deco; Noisy Brain 31 A feature of autoassociative memory is that it is content addressable; i.e. the information in the memory can be accessed if just the contents of the memory (or in part of the contents of the memory) are used. 0
Rolls & Deco; Noisy Brain 31 Autoassociation memory can be used as a short-term memory, in which iterative processing round the recurrent collateral connection loop keeps a representation active by continuing neuronal firing. 0
Rolls & Deco; Noisy Brain 31 The short-term memory reflected in continuing neuronal firing for several hundred milliseconds after a visual stimulus is removed, which is present in visual cortical areas such as inferior temporal visual cortex, is probably implemented in this way. 0
Rolls & Deco; Noisy Brain 31 Autoassociation memory appears to be used in a short-term memory in the prefrontal cortex. Temporal visual cortical areas have connections to the ventrolateral prefrontal cortex that help to implement short-term memory for visual stimuli. 0
Rolls & Deco; Noisy Brain 31 Parietal cortex has connections to dorsolateral prefrontal cortex for the short-term memory of spatial responses. 0
Rolls & Deco; Noisy Brain 31 Prefrontal short-term memories provide a mechanism that enables attention to be maintained through backprojections from the frontal cortex areas to the temporal and parietal areas that send connections to the prefrontal cortex. 0
Rolls & Deco; Noisy Brain 31 Attractor networks are implemented by the forward and backward connections between cortical areas. 0
Rolls & Deco; Noisy Brain 31 An autoassociation network with the rapid synaptic plasticity can learn each memory in one trial. Because of its one-shot rapid learning, and ability to complete, this type of network is well suited for episodic memory storage, in which each past episode must be stored and recalled later from fragments, and kept separate from other episodic memories. 0
Rolls & Deco; Noisy Brain 35 Perhaps the most important and useful property of autoassociative memories is that they complete an incomplete input vector, allowing recall of a whole memory from a small fraction of it. 4
Rolls & Deco; Noisy Brain 35 The memory recalled in response to a fragment is that stored in the memory that is closest in pattern similarity (as measured by the dot product, or correlation). 0
Rolls & Deco; Noisy Brain 35 Because the recall in associative memory is iterative and progressive, the recall can be perfect. 0
Rolls & Deco; Noisy Brain 36 If the network is modeled as 'integrate-and-fire' neurons, it can effectively relax into its recall state very rapidly, in one or two time constants of the synapses. This corresponds to perhaps 20 ms in the brain. 1
Rolls & Deco; Noisy Brain 36 One factor in the rapid dynamics of associative networks with brain-like 'integrate-and-fire' membrane and synaptic properties is that with ongoing spontaneous activity, some of the neurons in the network are close to threshold already, before the recall cue is applied, and hence some of the neurons are very quickly pushed by the recall cue into firing, perhaps within 1-2 ms. 0
Rolls & Deco; Noisy Brain 41 In a sparse distributed representation, a small proportion of the neurons its active at any one time to represent any one stimulus or event. 5
Rolls & Deco; Noisy Brain 42 One advantage of distributed encoding is that the similarity between two representations can be reflected by the correlation between the two patterns of activity that represent the different stimuli. 1
Rolls & Deco; Noisy Brain 42 Another advantage of distributed encoding is that the number of different stimuli that can be represented can be very large. 0
Rolls & Deco; Noisy Brain 43 Information can be read from the ensemble of neurons using a simple measure of the similarity of vectors, the correlation (or dot product). 1
Rolls & Deco; Noisy Brain 43 Vector similarity operations characterize the operation of many neuronal networks. 0
Rolls & Deco; Noisy Brain 43 Neurophysiological results show that both the ability to reflect similarity by vector correlation, and the utilization of exponential coding capacity, are properties of real neuronal networks in the brain. 0
Rolls & Deco; Noisy Brain 43 The brain codes stimuli in such a way that the code can be read off a simple dot product or correlation-related decoding. 0
Rolls & Deco; Noisy Brain 54 The spiking of typical cortical neurons is close to that of a Poisson process, i.e. for a given mean firing rate, the emission times of the spikes are approximately random and independent of each other. 11
Rolls & Deco; Noisy Brain 55 The time constants of the synapses are an important factor in determining how noise, including that from the stochastic firing of other neurons, affects the operation of the network. 1
Rolls & Deco; Noisy Brain 55 Whereas the AMPA and GABA receptors operate with relatively short time constants (in the order of 5 ms), the NMDA receptors have a much longer time constant (in the order of 100 ms), which dampens the effect of a high frequency noise and helps to promote stability and dampen oscillations in the cortical network. 0
Rolls & Deco; Noisy Brain 55 The neuronal membrane and dendritic time constants are in the order of 20 ms. Their effect on the temporal dynamics of the network is less than that of the synapses. 0
Rolls & Deco; Noisy Brain 55 The overall temporal properties of the network will be influenced by the synaptic and membrane time constants. 0
Rolls & Deco; Noisy Brain 55 Given that the time constants of GABA synapses is in the order of 10-20 ms, of AMPA receptors is 2-5 ms, and NMDA receptors is approximately 100 ms, oscillations are less likely to occur when the system is strongly activated so that the neurons are firing fast, and the voltage-independent NMDA receptors are contributing significantly to the currents in the system. 0
Rolls & Deco; Noisy Brain 57 Information theory provides the means for quantifying how much neurons communicate to other neurons, and thus provide a quantitative approach to fundamental questions about information processing in the brain. 2
Rolls & Deco; Noisy Brain 57 We must investigate to what extent the information provided by different cells is redundant or instead independent. 0
Rolls & Deco; Noisy Brain 58 Response of single neurons in the inferior temporal visual cortex to sets of objects and/or faces. 1
Rolls & Deco; Noisy Brain 61 Across a population of cells, the firing rate information provided by each neuron tends to be independent; i.e. the information increases approximately linearly with the number of neurons. 3
Rolls & Deco; Noisy Brain 61 Linear increase in information with the number of neurons has been found in many research studies, including for face and object representations in the inferior temporal visual cortex, for neurons responding to spatial view in the hippocampus, for neurons responding to head direction in the presubiculum, and for neurons responding to olfactory and taste stimuli in the orbitofrontal cortex. 0
Rolls & Deco; Noisy Brain 61 The number of stimuli that can be encoded rises exponentially as the number of neurons in the ensemble increases. 0
Rolls & Deco; Noisy Brain 62 The information in the firing rate across a population of neurons can be read moderately efficiently by decoding procedure as simple as a dot product. 1
Rolls & Deco; Noisy Brain 62 The dot product operation is the simplest type of processing that might be performed by neuron, as it involves taking a dot product of the incoming firing rates with the receiving synaptic weights to obtain the activation of the neuron. 0
Rolls & Deco; Noisy Brain 62 Dot product encoding ensures that the simple emergent properties of associative neuronal networks such as generalization, completion, and graceful degradation can be realized very naturally and simply. 0
Rolls & Deco; Noisy Brain 62 There is little additional information, to the great deal available in the firing rates, from any stimulus-dependent cross-correlations of synchronization that may be present. 0
Rolls & Deco; Noisy Brain 62 Stimulus dependent synchronization might only be useful for grouping different neuronal populations, and would not easily provide a solution to the binding problem in vision. 0
Rolls & Deco; Noisy Brain 62 The binding problem in vision may be solved by the presence of neurons that respond to combinations of features in a given spatial position with respect to each other. 0
Rolls & Deco; Noisy Brain 62 There is little additional information to the great deal available in the firing rates, from the order on which spikes arrive from different neurons. 0
Rolls & Deco; Noisy Brain 65 The spiking of neurons in the brain is almost random in time for a given mean rate, i.e. the spiking is approximately Poisson, and this randomness introduces noise into the system, which makes the system behave stochastically. 3
Rolls & Deco; Noisy Brain 65 The integrate-and-fire approach is a computationally realizable way to simulate the operation of neural networks when operating stochastically. 0
Rolls & Deco; Noisy Brain 65 In the simulation, each of the neurons show spiking that it's very much like that of neurons recorded in the brain. 0
Rolls & Deco; Noisy Brain 73 Autoassociation attractor systems have two types of stable fixed points: (1) a spontaneous state with a low firing rate, and (2) one or more attractor states in which the positive feedback implemented by the recurrent collateral connections maintain a high firing rate. 8
Rolls & Deco; Noisy Brain 73 The high firing rate attractor state is sometimes referred to as the persistent state, because the high firing rate persists to maintain a set of neurons active, which might implement a short-term memory. 0
Rolls & Deco; Noisy Brain 73 The stable points in the system can be visualized in an energy landscape. 0
Rolls & Deco; Noisy Brain 73 The area in the energy landscape within which the system will move to a stable attractor state is call the basin of attraction. 0
Rolls & Deco; Noisy Brain 74 The attractor dynamics can be pictured by energy landscapes, which indicate the basins of attraction by valleys, and the attractor states or fixed points by the bottom of the valleys. 1
Rolls & Deco; Noisy Brain 74 The stability of an attractor is characterized by the average time in which the system stays in the basin of attraction under the influence of noise. 0
Rolls & Deco; Noisy Brain 74 Noise provokes transitions to other attractor states. 0
Rolls & Deco; Noisy Brain 74 Two factors determined the stability: (1) the depths of the attractors, and (2) high noise will make it more likely that the system will jump over an energy boundary from one state to another. 0
Rolls & Deco; Noisy Brain 74 The noise could arise not only from the probabilistic spiking of the neurons, but also from any other source of noise in the brain or environment, including the effects that is distracting stimuli. 0
Rolls & Deco; Noisy Brain 78 Neurons are held close to (just slightly below) their firing threshold so that any incoming input can rapidly cause sufficient further depolarization to produce a spike. 4
Rolls & Deco; Noisy Brain 78 It is this ability to respond rapidly to an input, rather than than having to charge up the cell membrane from the resting potential to the threshold, that enables neuronal networks in the brain, including attractor networks, to operate and retrieve information so rapidly. 0
Rolls & Deco; Noisy Brain 78 Charging up to cell membrane from the resting potential to the threshold is a slow process determined by the time constant of the neuron and influenced by the synapses. 0
Rolls & Deco; Noisy Brain 78 Spike trains are essentially Poisson-like because the cell potential hovers noisily close to the threshold for firing, the noise being generated in part by the Poisson-like firing of the other neurons in the network. 0
Rolls & Deco; Noisy Brain 78 Noise and spontaneous firing help to ensure that when a stimulus arrives, there are always some neurons very close to threshold that respond rapidly, and then communicate their firing to other neurons through the modified synaptic weights, so that an attractor process can take place very rapidly. 0
Rolls & Deco; Noisy Brain 79 Noise sources in the brain arise from effects that make firing properties of neurons very close to Poisson. 1
Rolls & Deco; Noisy Brain 79 If the system has a finite number of neurons, then the Poisson or semi-random effects will not average out, and the average firing rate of the population will fluctuate statistically. 0
Rolls & Deco; Noisy Brain 80 Signal-to-noise ratio can be measured in a network by the average firing rate change produced by the signal in the neuronal population divided by the standard deviation of the firing rates. 1
Rolls & Deco; Noisy Brain 80 Noise inherent in brain activity has a number of advantages by making the dynamics stochastic, which allows for many remarkable features of the brain, including creativity, probabilistic decision making, stochastic resonance, unpredictability, conflict resolution, symmetry breaking, allocation to discrete categories, and many of the important properties. 0
Rolls & Deco; Noisy Brain 81 Statistical fluctuations are not just a concomitant of neural processing, but can play an important and unique role in cortical function. 1
Rolls & Deco; Noisy Brain 81 Computations can be performed through stochastic dynamical effects, including the role of noise in enabling probabilistic jumping across the barrier in the energy landscape describing the flow of the dynamics in attractor networks. 0
Rolls & Deco; Noisy Brain 81 Stochastic dynamical effects help to understand short-term memory, attention, long-term memory, and decision-making in the brain, and to establish a link between neuronal firing variability in probabilistic behavior. 0
Rolls & Deco; Noisy Brain 82 Mean-field based theory -- starting from the mathematical models of biologically realistic single neurons (i.e. spiking neurons), one can derive models that describe the joint activity of pools (or populations) of equivalent neurons. The computational unit is a neuronal pool, population, or assembly. 1
Rolls & Deco; Noisy Brain 91 Short-Term Memory and Stochastic Dynamics 9
Rolls & Deco; Noisy Brain 91 Cortical short-term memory systems and attractor networks 0
Rolls & Deco; Noisy Brain 91 There are a number of different short-term memory systems, each implemented in a different cortical area. 0
Rolls & Deco; Noisy Brain 91 The short-term memories may operate as autoassociative attractor networks. 0
Rolls & Deco; Noisy Brain 91 The short-term memory is implemented by subpopulations of neurons that show maintained activity in a delay., while a stimulus or event is being remembered. 0
Rolls & Deco; Noisy Brain 91 The autoassociation could be implemented by associatively modifiable synapses between connected pyramidal cells within an area, or by the forward and backward connections between adjacent cortical areas in a hierarchy. 0
Rolls & Deco; Noisy Brain 91 One short-term memory system is in the dorsolateral prefrontal cortex, area 46. This is involved in remembering the locations of spatial responses. 0
Rolls & Deco; Noisy Brain 91 There is another short-term memory system in a more dorsal and posterior part of the prefrontal cortex involved in remembering the position in visual space to which an eye movement (a saccade) should be made. In this case the short-term memory function is topographically organized. 0
Rolls & Deco; Noisy Brain 91 Local cortical connections between nearby pyramidal cells implement an attractor network. 0
Rolls & Deco; Noisy Brain 92 Another short-term memory system is implemented in the inferior temporal visual cortex. 1
Rolls & Deco; Noisy Brain 92 Some neurons respond more to a novel than a familiar visual stimulus. 0
Rolls & Deco; Noisy Brain 93 A possible cortical model of a topographically organized set of attractor networks in the prefrontal cortex that could be used to remember the positions to which saccades should be made. (Diagram) 1
Rolls & Deco; Noisy Brain 93 Short-term memory is also found in the inferior frontal convexity cortex in a region connected to the ventral temporal cortex. 0
Rolls & Deco; Noisy Brain 94 Prefrontal cortex short-term memory networks, and their relation to temporal and parietal perceptual networks. 1
Rolls & Deco; Noisy Brain 94 A common way that the brain uses to implement a short-term memory is to maintain the firing of neurons during a short memory period after the end of a stimulus. 0
Rolls & Deco; Noisy Brain 94 In the inferior temporal cortex the firing of neurons for short-term memory may be maintained for a few hundred milliseconds. In more ventral temporal cortical areas such as the entorhinal cortex the firing may be maintained for longer periods. In the prefrontal cortex the firing may be maintained even for tens of seconds. 0
Rolls & Deco; Noisy Brain 94 In the dorsolateral and inferior convexity prefrontal cortex the firing of the neurons may be related to the memory of spatial responses or objects. 0
Rolls & Deco; Noisy Brain 94 The firing of neurons may be maintained by the operation of associatively modified recurrent collateral connections between nearby pyramidal cells producing attractor states in autoassociative networks. 0
Rolls & Deco; Noisy Brain 94 For short-term memory to the maintained during periods in which new stimuli are to be perceived, and must be separate networks for perceptual and short-term memory functions. 0
Rolls & Deco; Noisy Brain 94 Two coupled networks, one in the inferior temporal visual cortex for perceptual functions, and another in the prefrontal cortex for maintaining the short-term memory during intervening stimuli, provide a precise model of the interaction of perceptual and short-term memory systems. 0
Rolls & Deco; Noisy Brain 94 A prefrontal cortex attractor (autoassociation) network could be triggered by a sample visual stimulus represented in the inferior temporal visual cortex and keep this attractor active during a memory interval in which intervening stimuli are shown. 0
Rolls & Deco; Noisy Brain 94 For ongoing perception to occur unhindered, implemented by posterior cortex (parietal and temporal lobe) networks, there must be a separate set of modules that is capable of maintaining a representation over intervening stimuli. 0
Rolls & Deco; Noisy Brain 95 A short-term memory autoassociation network in the prefrontal cortex could hold active a working memory representation by maintaining its firing in an attractor state. (diagram) 1
Rolls & Deco; Noisy Brain 95 The ability of the dorsolateral prefrontal cortex to provide multiple separate short-term attractor memories could provide the basis for its function in planning. 0
Rolls & Deco; Noisy Brain 95 Neuronal activity in the inferior temporal visual cortex (IT) is driven by each new incoming visual stimulus, whereas in the prefrontal cortex, neurons start to fire when the sample stimulus is shown, and continue the firing that represents the sample stimulus, even when potential match stimuli are being shown. 0
Rolls & Deco; Noisy Brain 95 Computer simulation shows that a crucial parameter is the strength of the intermodular connections between the prefrontal cortex and the posterior cortex (temporal or parietal cortex). 0
Rolls & Deco; Noisy Brain 97 Computer simulation shows that the operation of the prefrontal cortex in short-term memory and its relation to posterior perceptual networks, can be understood by the interaction of two weakly coupled attractor networks. 2
Rolls & Deco; Noisy Brain 99 Inputs from the parietal cortex, involved in spatial computation, project more to the dorsolateral prefrontal cortex. 2
Rolls & Deco; Noisy Brain 99 Inputs from the temporal lobe visual cortex, involved in time dictations about objects, project ventrolaterally to the prefrontal cortex. 0
Rolls & Deco; Noisy Brain 99 It could be that the dorsolateral prefrontal cortex is especially involved in spatial short-term memory, whereas the inferior convexity prefrontal cortex more in object short-term memory. 0
Rolls & Deco; Noisy Brain 99 The organization-by-stimulus-domain hypothesis holds that the spatial ('where') working memory is supported by the dorsolateral PFC, BA 46/9, in the middle frontal gyrus (MFG). 0
Rolls & Deco; Noisy Brain 99 Object ('what') working memory is supported by the ventrolateral PFC on the lateral/inferior connectivity, BA 45. 0
Rolls & Deco; Noisy Brain 99 Some mixing of the spatial and object representations in the lateral prefrontal cortex is needed in order for some short-term memory tasks to be resolved, and thus total segregation of 'what' and 'where' processing in the lateral prefrontal cortex should not be is expected. 0
Rolls & Deco; Noisy Brain