The tempotron: a neuron that learns spike timing–based decisions

By Robert Gütig, Haim Sompolinsky. Nature Neuroscience, 2006

844 citations recorded by the source corpus.

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

View Paper: https://www.semanticscholar.org/paper/eaba315d5f972c059ab2d0315f2aeada27c1becd

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

Sparseness and expansion in sensory representations. (2014, background): “However, a careful analysis reveals that, for large random patterns, capacity is improved by sparseness of their class membership but is unaffected by sparseness of the inputs to the classifier (Gardner, 1988; Gütig and Sompolinsky, 2006).”
Neuronal Synchrony in Complex-Valued Deep Networks (2013, methodology): “The DUBM is a principled probabilistic framework, within which a complexvalued extension of Hebbian learning can be derived, with 10Consider for instance the tempotron neuron model (Gütig & Sompolinsky, 2006), which learns to recognize spike patterns.”
Incorporating Learnable Membrane Time Constant to Enhance Learning of Spiking Neural Networks (2020, methodology): “Supervised learning of SNNs SpikeProp [5] was the first supervised learning method for SNNs based on backpropagation, which used a linear approximation to overcome the non-differentiable threshold-triggered firing mechanism of SNNs. Subsequent works included Tempotron [18], ReSuMe [45], and SPAN [40”
Simulation of networks of spiking neurons: A review of tools and strategies (2006, positive): “For example, consider the following IF model (described for example in Gütig and Sompolinsky (2006)): where V ( t ) is the membrane potential, V rest is the rest potential, ω i is the synaptic weight of synapse i , t i are the timings of the spikes coming from synapse i , and K ( t − t i ) = exp (…”
SpikingJelly: An open-source machine learning infrastructure platform for spike-based intelligence (2023, neutral): “Primitive supervised learning methods including SpikeProp ( 39 ), Tempotron ( 40 ), ReSume ( 41 ), and SPAN ( 42 ) achieve higher performance than biologically plausible unsupervised methods.”
State-dependent computations: spatiotemporal processing in cortical networks (2009, background): “Traditional artificial neural network learning rules do not capture information that is contained in previous time steps — that is, information encoded in the temporal pattern of active states; however, some supervised-learning rules have also proved effective in allowing spiking read-out neurons to”
SuperSpike: Supervised Learning in Multilayer Spiking Neural Networks (2017, background): “2b; Gütig and Sompolinsky (2006)).”
The remarkable robustness of surrogate gradient learning for instilling complex function in spiking neural networks (2020, methodology): “…a μ i for each input μ were depending on the readout configuration either given by the sum over all time steps a μ i = (cid:11) n (cid:5) U (out) i (cid:6) (cid:5) (cid:6) [ n ] or defined as the maximum a μ i = max n U (out) i [ n ] , a notion inspired by the Tempotron (Gütig & Sompolinsky, 2006).”
Plasticity of Cortical Excitatory-Inhibitory Balance (2015, methodology): “This has made STDP useful for theoretical studies of synaptic plasticity in neural networks (Clopath et al. 2010, Gjorgjieva et al. 2011, Gütig & Sompolinsky 2006, Pfister & Gerstner 2006, Song et al. 2000, Vogels et al. 2011, Yao et al. 2004), as pre- and postsynaptic spike pairs can be…”
Introduction to spiking neural networks: Information processing, learning and applications. (2011, background): “Up to now many other interesting algorithms for supervised learning in spiking networks have been proposed, including: Sougne (2001), Pfister and others (2006), Gütig and Sompolinsky (2006), Schrauwen and Campenhout (2006).”
Deep Learning With Spiking Neurons: Opportunities and Challenges (2018, background): “Several promising, but not necessarily deep, temporal coding and learning schemes have been proposed (e.g., Gütig and Sompolinsky, 2006; Tapson et al., 2013; Lagorce and Benosman, 2015), but no applications at the scale that conventional deep learning is addressing have been demonstrated, yet.”
Evolving spiking neural networks for audiovisual information processing (2010, background): “This phenomenon has been experimentally verified in a series of independent works (see Abeles & Gat, 2001) and computational models explored the storage and learning capabilities of this theory (Bienenstock, 1995; Gutig & Sompolinsky, 2006).”
Neuromorphic Electronic Circuits for Building Autonomous Cognitive Systems (2014, methodology): “This is the condition for linearity, under which the circuit starts to behave as a first order low-pass filter, as described in Section II.”