Passive Nonlinear Dendritic Interactions as a Computational Resource in Spiking Neural Networks

Area of Science:

• Computational Neuroscience
• Artificial Neural Networks
• Biophysics

Background:

• Dendritic nonlinearities are crucial for neural computation but often ignored in large-scale network models.
• Existing frameworks like the Neural Engineering Framework (NEF) typically assume linear summation of postsynaptic currents.
• This limits the ability to model complex neuronal behaviors and computations.

Purpose of the Study:

• To extend the Neural Engineering Framework (NEF) to incorporate nonlinear dendritic computations.
• To enable the construction of spiking neural networks that adhere to Dale's principle and utilize nonlinear synapses.
• To investigate the computational power of nonlinear dendritic models within a large-scale network framework.

Main Methods:

• Developed extensions to the NEF to support nonlinear conductance-based synapses and Dale's principle.
• Applied these extensions to a two-compartment Leaky Integrate-and-Fire (LIF) neuron model.
• Analyzed the network's ability to approximate multivariate functions using nonlinear postsynaptic currents.

Main Results:

• Successfully integrated neuron models with input-dependent nonlinearities into the NEF without compromising network function.
• Demonstrated that nonlinear postsynaptic currents can compute diverse multivariate functions, including Euclidean norm and controlled shunting.
• Showcased that a single layer of nonlinear two-compartment LIF neurons can achieve function approximation accuracy comparable to or exceeding two-layer linear networks.

Conclusions:

• The extended NEF effectively incorporates nonlinear dendritic processing into large-scale spiking neural networks.
• Nonlinear dendritic computations offer a powerful mechanism for enhancing the functional capabilities of artificial neural networks.
• This approach provides a more biologically plausible and computationally efficient method for building complex neural models.