Bayesian Computation Emerges in Generic Cortical Microcircuits through Spike-Timing-Dependent Plasticity

By Bernhard Nessler, Michael Pfeiffer, Lars Buesing, Wolfgang Maass. 2013

310 citations recorded by the source corpus.

The principles by which networks of neurons compute, and how spike-timing dependent plasticity (STDP) of synaptic weights generates and maintains their computational function, are unknown. Preceding work has shown that soft winner-take-all (WTA) circuits, where pyramidal neurons inhibit each other via interneurons, are a common motif of cortical microcircuits. We show through theoretical analysis and computer simulations that Bayesian computation is induced in these network motifs through STDP in combination with activity-dependent changes in the excitability of neurons. The fundamental components of this emergent Bayesian computation are priors that result from adaptation of neuronal excitability and implicit generative models for hidden causes that are created in the synaptic weights through STDP. In fact, a surprising result is that STDP is able to approximate a powerful principle for fitting such implicit generative models to high-dimensional spike inputs: Expectation Maximization. Our results suggest that the experimentally observed spontaneous activity and trial-to-trial variability of cortical neurons are essential features of their information processing capability, since their functional role is to represent probability distributions rather than static neural codes. Furthermore it suggests networks of Bayesian computation modules as a new model for distributed information processing in the cortex.

Subfields: Cellular and Molecular Neuroscience, Cognitive Neuroscience, Electrical and Electronic Engineering

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cellular and Molecular Neuroscience

In Cognitive Neuroscience

In Electrical and Electronic Engineering

Citations in the corpus

Phased LSTM: Accelerating Recurrent Network Training for Long or Event-based Sequences (2016, background): “An inspiration for using oscillations in recurrent networks comes from computational neuroscience [3], where rhythms have been shown to play important roles for synchronization and plasticity [22].”
Real-time classification and sensor fusion with a spiking deep belief network (2013, background): “Nessler et al. (2013) have shown that learning via STDP in cortical microcircuits can lead to the emergence of Bayesian computation for the detection of hidden causes of inputs.”
Spin-transfer torque magnetic memory as a stochastic memristive synapse (2014, unknown): “The learning rule is also reminiscent of the simplified STDP proposed for computational neuroscience studies in [59], but in a binary and probabilistic version.”
Unsupervised learning of digit recognition using spike-timing-dependent plasticity (2015, methodology): “The second STDP rule uses an exponential weight dependence (Nessler et al., 2013; Querlioz et al., 2013) to compute the weight change where β determines the strength of the weight dependence.”
Noise as a Resource for Computation and Learning in Networks of Spiking Neurons (2014, methodology): “This figure was previously published in [15] under the CCAL.”
Dendritic cortical microcircuits approximate the backpropagation algorithm (2018, background): “, 2011; Froemke, 2015) and competition (Masquelier and Thorpe, 2007; Nessler et al., 2013).”
Computational models of long term plasticity and memory (2017, unknown): “The role of STDP in learning has also been investigated in many computational studies (see e.g. [Gerstner et al., 1996, Legenstein et al., 2005b, Gutig and Sompolinsky, 2006, Izhikevich, 2007, Legenstein et al., 2008, Nessler et al., 2013, Pecevski and Maass, 2016]).”
Emergence of Dynamic Memory Traces in Cortical Microcircuit Models through STDP (2013, methodology): “Furthermore it is shown in (Nessler et al., 2013), see Figure 8 of that article and the surrounding discussion, that the dependence of weight updates on the current weight in (5) is qualitatively consistent with published experimental data (see also the discussion in section 2.”
Spiking Deep Networks with LIF Neurons (2015, methodology): “Networks could also be trained ofﬂine as described here and then ﬁne-tuned online using an STDP rule, such as the one described in [22], to help further reduce errors associated with converting from rate-based to spike-based networks, while avoiding difﬁculties with training a network in spiking…”
Stochastic variational learning in recurrent spiking networks (2014, methodology): “These interesting properties of the variational approximation have been explored by a diversity of models (Dayan, 2000; Friston and Stephan, 2007; Jimenez Rezende et al., 2011; Brea et al., 2013; Nessler et al., 2013).”
Incorporating Learnable Membrane Time Constant to Enhance Learning of Spiking Neural Networks (2020, methodology): “Existing approaches exploited the self-organization principle [56, 11, 29], and STDP-based expectation-maximization algorithm [43, 17].”
Memory and Information Processing in Neuromorphic Systems (2015, background): “, the neuron’s membrane potential) [96], [97], and/or the current value of the synaptic weights [98], [99].”
Deep Learning in Spiking Neural Networks (2018, methodology): “They used unsupervised learning rules introduced by [131], [90], [132] to train the SNN for the Modified National Institute of Standards and Technology (MNIST) dataset [133] digit recognition with the best performance of 95%.”
Neuromodulated Spike-Timing-Dependent Plasticity, and Theory of Three-Factor Learning Rules (2016, background): “For example, for the learning of binary decisions in populations of spiking neurons, a neuromodulatory signal proportional to the population activity has been suggested (Urbanczik and Senn, 2009; Friedrich et al., 2011).”
Deep Learning With Spiking Neurons: Opportunities and Challenges (2018, background): “In competitive networks, STDP is capable of solving unsupervised learning tasks such as clustering (Masquelier et al., 2009; Nessler et al., 2013).”