Embedding low-dimensional continuous dynamical systems in recurrently connected spiking neural networks
Despite recent advances in training recurrently connected firing-rate networks, the application of supervised learning algorithms to biologically plausible recurrently connected spiking neural networks remains a challenge. Such models, when trained to directly replicate neural data, hold great promise as powerful tools for understanding dynamic computation in biologically realistic neural circuits. In this talk I will discuss our progress in the training of recurrently connected spiking networks, the application of our training framework to neural population data and a novel interpretation of continuous neural signals that arises within the context of these models.
Extending the iterative supervised learning algorithm of Sussillo & Abbott , we have made several critical observations about the conditions necessary for successfully training recurrent spiking networks. Due to their impoverished short-term memory, multiple signals that form a “dynamically complete” basis must be trained simultaneously for successful training. I will illustrate this by showing a variety of examples of spiking neural networks replicating the dynamics of both autonomous and non-autonomous linear and non-linear continuous dynamical systems. Additionally, I will discuss recent efforts to incorporate a variety of network optimization constraints such that the learned connectivity matrices obey common constraints of biological networks, including sparsity and Dale’s Law. Finally, I will discuss our efforts to fit spiking models to population data from the isolated nervous system of the leech.
Once trained, our models can be viewed as a low-dimensional, continuous dynamical system - traditionally modeled with firing-rate networks - embedded in a high-dimensional, spiking dynamical system. In light of this view, I will present a novel interpretation of firing-rate models and smoothly varying signals in general. Traditionally a continuous neural signal modeled as a “firing-rate unit” was a simplified representation of a pool of identical but noisy spiking neurons. In our formulation, each continuous neural signal represents an overlapping population of spiking neurons and is thus more akin to the multiple, continuous population trajectories one would uncover from experimental data via dimensionality reduction. By allowing these continuous signals to be constructed from overlapping pools of spiking neurons, our framework requires far fewer spiking neurons to arrive at the equivalent, traditional rate description.