580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can simplified adaptive exponential integrate-and-fire (AdEx) models reproduce diverse neuronal firing patterns and support efficient large-scale neural simulations?
This research theme focuses on developing and validating computationally efficient neuron models based on the adaptive exponential integrate-and-fire (AdEx) framework, aiming to balance biological realism with simulation efficiency. These simplified point-neuron models capture complex electrophysiological behaviors such as spike-frequency adaptation, bursting, and varied firing types, enabling classification of neuron types and incorporation into large-scale network models. The theme addresses methodological advances in model parameterization, hardware implementation, and mean-field approximations to facilitate biologically plausible yet computationally tractable simulations of cortical dynamics.
Key finding: Introduces a suite of Generalized Leaky Integrate-and-Fire (GLIF) models, including adaptive mechanisms, fitted to electrophysiological data from 771 neurons. Results demonstrate that increasing model complexity improves... Read more
Key finding: Presents an efficient digital hardware implementation of the AdEx neuron using a CORDIC algorithm to approximate the exponential nonlinearity. The hardware is demonstrated on FPGA to reproduce key neuronal behaviors and... Read more
Key finding: Develops a mean-field model of cortical network dynamics built from conductance-based AdEx neurons with spike-frequency adaptation. The model accurately predicts spontaneous and stimulus-evoked population activity in networks... Read more
Key finding: Adapts a semi-analytical mean-field approach to conductance-based networks of adaptive quadratic integrate-and-fire (aQIF) neurons, incorporating excitatory and inhibitory populations and adaptation. The mean-field accurately... Read more
2. What are the mathematical and computational strategies for refining adaptive integrate-and-fire neuron models to account for spike-frequency adaptation and complex firing dynamics?
This theme investigates analytical and computational methods to incorporate spike-frequency adaptation and multiple timescale dynamics into integrate-and-fire neuron models. Approaches involve fractional calculus, refractory density formalisms, and population rate modeling to understand and reproduce phenomena such as power-law spike adaptation, bursting, and transient dynamics. These methods provide bridges from microscopic ion channel behavior to macroscopic population activity and propose reduced-dimensional firing rate models capturing the essential features of adaptive neurons.
Key finding: Introduces a fractional derivative leaky integrate-and-fire model incorporating a fractional exponent (0 < α ≤ 1) to capture long-range temporal correlations in membrane voltage due to slowly adapting ionic conductances. The... Read more
Key finding: Reviews recent advancements in firing rate models derived from detailed integrate-and-fire neuron descriptions with adaptation, using refractory density (RD) methods. Demonstrates that RD enables exact population descriptions... Read more
Key finding: Proposes a firing rate (FR) model for populations of adaptive leaky integrate-and-fire neurons by transforming voltage-dependent adaptive conductances into spike and subsequently rate-dependent variables. The FR model... Read more
Key finding: Generalizes the conductance-based refractory density (CBRD) framework, originally for regular spiking neurons, to populations capable of bursting dynamics with Hodgkin-Huxley-type ion channels. The approach reduces... Read more
3. How can network connectivity patterns and numerical methods be used to model and analyze spike timing, oscillatory dynamics, and synchronization phenomena in adaptive integrate-and-fire neuronal networks?
This theme explores the impact of different network architectures, spike timing-dependent plasticity (STDP), and numerical integration schemes on the collective dynamics of adaptive integrate-and-fire neural networks. It investigates phenomena such as chimera states, firing rate propagation, population oscillations influenced by adaptation, and methods for accurate and efficient simulation of spike timing. Studies include applying surrogate gradients to train LIF networks, modeling synaptically coupled layers, and analyzing the emergence and modulation of collective oscillations and phase-amplitude coupling in excitatory/inhibitory populations with adaptation.
Key finding: Develops an integrate-and-fire spiking neural network incorporating spike-timing dependent plasticity (STDP) that reproduces multiple conditioning protocols inducing cortical plasticity observed in primate experiments. The... Read more
Key finding: Demonstrates using Hodgkin-Huxley neuron networks that white noise applied to the input layer can generate synchronous firing patterns that build up and propagate through multilayer feed-forward architectures, enabling stable... Read more
Key finding: Investigates chimera states in LIF networks with nonlocal and reflecting connectivities, demonstrating emergence of spatial patterns with alternating active and near-threshold domains. Reflecting connectivity, motivated by... Read more
Key finding: Develops and mathematically analyzes modified Runge-Kutta integration schemes that accurately handle reset discontinuities in integrate-and-fire neuron networks, including a new fourth-order method. By interpolating spike... Read more
Key finding: Introduces a method for integrating LIF neurons into machine learning architectures using surrogate gradients to enable backpropagation through non-differentiable spike functions. Demonstrates that tuning the leak term... Read more