Reproduce “Point Neurons with Conductance-Based Synapses in the Neural Engineering Framework” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Andreas Stöckel, Aaron R. Voelker, Chris Eliasmith
Researchers at the Centre for Theoretical Neuroscience at the University of Waterloo integrated biologically realistic conductance-based synapses into the Neural Engineering Framework (NEF), demonstrating that a direct translation method is effective for feed-forward signal transmission but fails for complex recurrent computations like integration due to inherent nonlinearities and altered neuronal tuning curves.
Key insights
- Conductance-based inhibition exerts a significantly stronger shunting effect than current-based inhibition, leading to a pronounced reduction in neuronal firing rates, especially with spiking input.
- While the naïve translation works adequately for feed-forward computations (e.g., identity function), it fundamentally breaks down for recurrent dynamics, as observed in the integrator task.
- Conductance-based neurons exhibit a "delayed spike onset" and systematically shifted tuning curves compared to current-based neurons, leading to inaccuracies in value representation crucial for recurrent system performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.