Efficient codes and balanced networks

Denève, Sophie; Machens, Christian K.

doi:10.1038/nn.4243

Cited by 434 publications

(562 citation statements)

References 77 publications

Supporting

Mentioning

510

Contrasting

Order By: Relevance

“…Nevertheless, a decrease in GABAergic neurotransmission has been observed in MS. 33 In fact, previous neural mass model-based studies have shown that even a small loss of interneurons, and thus a small drop in inhibitory activity, can induce a very large functional effect on hubs and the entire brain network. 34 However, given that excitatory neurons are most likely also affected by MS, it seems that a lower inhibitory activity could not be the only mechanism that underlies functional brain changes. For instance, a late and possibly ineffective attempt at beneficial functional reorganization cannot be excluded as an explanation for our results at this time.…”

Section: Structural Brain Measures White Matter (Wm) Lesions Werementioning

confidence: 99%

Increased connectivity of hub networks and cognitive impairment in multiple sclerosis

et al. 2017

View full text Add to dashboard Cite

Objective: To investigate default-mode network (DMN) and frontoparietal network (FPN) dysfunction in cognitively impaired (CI) patients with multiple sclerosis (MS) because these networks strongly relate to cognition and contain most of the hubs of the brain.Methods: Resting-state fMRI and neuropsychological assessments were performed in 322 patients with MS and 96 healthy controls (HCs). Patients with MS were classified as CI (z score , 22.0 on at least 2 tests; n 5 87), mildly cognitively impaired (z score , 21.5 on at least 2 tests and not CI; n 5 65), and cognitively preserved (CP; n 5 180). Within-network connectivity, connectivity with the rest of the brain, and between-network connectivity were calculated and compared between groups. Connectivity values were normalized for individual means and SDs.Results: Only in CI, both the DMN and FPN showed increased connectivity with the rest of the brain compared to HCs and CP, with no change in within-or between-network connectivity. Regionally, this increased connectivity was driven by the inferior parietal, posterior cingulate, and angular gyri. Increased connectivity with the rest of the brain correlated with worse cognitive performance, namely attention for the FPN as well as information processing speed and working memory for both networks.Conclusions: In CI patients with MS, the DMN and FPN showed increased connectivity with the rest of the brain, while normal within-and between-network connectivity levels were maintained. These findings indicate that cognitive impairment in MS features disturbed communication of hub-rich networks, but only with the more peripheral (i.e., nonhub) regions of the brain. Patients with multiple sclerosis (MS) commonly experience cognitive decline, which is most likely driven by functional network changes. 1-3 Across neurologic disorders, changes in connectivity of especially the default-mode network (DMN) and frontoparietal network (FPN) have been linked to cognitive deficits. [4][5][6] This preferential relationship might be explained by the fact that these networks contain the majority of highly connected regions, commonly described as functional hubs. 7,8 In fact, such hub regions are essential for optimal cognitive function because they ensure efficient integration of information between different brain regions. 9 Previous studies have reported both increased 1,10,11 and decreased 3,12,13 global connectivity levels of the DMN and FPN in relation to cognitive dysfunction in MS. This creates confusion about how these functional connectivity changes, either increased or decreased, may actually influence cognition. 14 Apart from the directionality of these changes, it remains unclear whether these effects are due specifically to changes in within-network connectivity, connectivity with the rest of the brain, or changes in between-network connectivity.

show abstract

Section: Structural Brain Measures White Matter (Wm) Lesions Werementioning

confidence: 99%

Increased connectivity of hub networks and cognitive impairment in multiple sclerosis

et al. 2017

View full text Add to dashboard Cite

show abstract

“…Increased conductance can reduce gain whilst making the membrane faster, extending bandwidth, and ensuring that action potentials occur more reliably. Balanced excitatory and inhibitory synaptic currents, which occur in many cortical neurons in vivo (reviewed in [48]), demonstrate the impact of synaptic conductances on energy efficiency. Computational modelling shows that single compartments receiving balanced excitatory and inhibitory synaptic currents achieve similar information rates to those receiving only excitatory inputs (or balanced excitatory and inhibitory conductances) but do so with fewer action potentials and, therefore, lower energy consumption [47].…”

Section: Synaptic Inputsmentioning

confidence: 99%

“…Graded potentials encode more information per unit time that pulsatile codes, and the conversion from graded to pulsatile produces information loss accompanied by a drop in energy efficiency [8,49,50]. Computational models show that information loss is due to increased intrinsic noise and non-linearity produced by the channels generating the action potential, as well as the duration of the action potentials themselves, which obscure the underlying graded signal [48]. Reduced efficiency is a consequence of information loss coupled with the cost of generating action potentials.…”

Section: Synaptic Inputsmentioning

confidence: 99%

Neuronal energy consumption: biophysics, efficiency and evolution

Niven

2016

Current Opinion in Neurobiology

119

View full text Add to dashboard Cite

Electrical and chemical signaling within and between neurons consumes energy. Recent studies have sought to refine our understanding of the processes that consume energy and their relationship to information processing by coupling experiments with computational models and energy budgets. These studies have produced insights into both how neurons and neural circuits function, and why they evolved to function in the way they do. Introduction Neurons consume energy. Appreciating that they do so is essential for understanding and interpreting the function and evolution of neurons, neural circuits and, ultimately, whole brains. Yet we must go beyond mere appreciation by relating specific molecular components and processes to the energy they consume and the work they contribute to processing information and generating behavior. This permits determination of both 'how' and 'why' processes consume energy, and an understanding of the key trade-offs that have influenced neural evolution (reviewed in [1,2]). Although neuronal energy consumption has been studied for over 80 years [3,4], conceptual and methodological breakthroughs [5][6][7][8][9] have prompted renewed interest in the causes and consequences of neuronal energy consumption over the last ~20 years. Here I review this recent progress in our understanding of how the physiology and anatomy of neurons and neural circuits reflect fundamental relationships between energy consumption, biophysics and performance. Addresses

show abstract

“…While some of these studies have been performed using spiking networks, they still use effectively a rate-based approach in which a given input activity vector is interpreted as the firing rate of a set of input neurons (Eliasmith et al., 2012; Diehl & Cook, 2015; Guergiuev et al., 2016; Neftci et al., 2016; Mesnard, Gerstner, & Brea, 2016). While this approach is appealing because it can often be related directly to equivalent rate-based models with stationary neuronal transfer functions, it also largely ignores the idea that individual spike timing may carry additional information that could be crucial for efficient coding (Thalmeier, Uhlmann, Kappen, & Memmesheimer, 2016; Denève & Machens, 2016; Abbott, DePasquale, & Memmesheimer, 2016; Brendel, Bourdoukan, Vertechi, Machens, & Denéve, 2017) and fast computation (Thorpe, Fize, & Marlot, 1996; Gollisch & Meister, 2008). …”

Section: Introductionmentioning

confidence: 99%

SuperSpike: Supervised Learning in Multilayer Spiking Neural Networks

2018

View full text Add to dashboard Cite

A vast majority of computation in the brain is performed by spiking neural networks. Despite the ubiquity of such spiking, we currently lack an understanding of how biological spiking neural circuits learn and compute in vivo, as well as how we can instantiate such capabilities in artificial spiking circuits in silico. Here we revisit the problem of supervised learning in temporally coding multilayer spiking neural networks. First, by using a surrogate gradient approach, we derive SuperSpike, a nonlinear voltage-based three-factor learning rule capable of training multilayer networks of deterministic integrate-and-fire neurons to perform nonlinear computations on spatiotemporal spike patterns. Second, inspired by recent results on feedback alignment, we compare the performance of our learning rule under different credit assignment strategies for propagating output errors to hidden units. Specifically, we test uniform, symmetric, and random feedback, finding that simpler tasks can be solved with any type of feedback, while more complex tasks require symmetric feedback. In summary, our results open the door to obtaining a better scientific understanding of learning and computation in spiking neural networks by advancing our ability to train them to solve nonlinear problems involving transformations between different spatiotemporal spike time patterns.

show abstract

Efficient codes and balanced networks

Cited by 434 publications

References 77 publications

Increased connectivity of hub networks and cognitive impairment in multiple sclerosis

Increased connectivity of hub networks and cognitive impairment in multiple sclerosis

Neuronal energy consumption: biophysics, efficiency and evolution

SuperSpike: Supervised Learning in Multilayer Spiking Neural Networks

Contact Info

Product

Resources

About