Reproduce “Mixed-signal implementation of feedback-control optimizer for single-layer Spiking Neural Networks” 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
Jonathan Haag, Christian Metzner, Dmitrii Zendrikov, Giacomo Indiveri, Benjamin Grewe, Chiara De Luca, Matteo Saponati
The authors test feedback-control learning on a mixed-signal DYNAP-SE processor. A single-layer spiking network runs forward dynamics on the chip, while controller neurons guide local weight changes and host software calculates updates and reconfigures synapses. On nonlinear three-class Yin–Yang, test accuracy is 67%, versus 63% in simulation and 66% for a prior-work single-layer ANN; updates are not native to silicon.
Key insights
- Rigorous calibration of neuron and synapse parameters, including input/target spike generation and controller biases, is crucial for stable and reliable operation of learning rules on mixed-signal neuromorphic hardware.
- Employing population coding with multiple neurons per unit effectively mitigates the impact of analog device mismatch and signal noise, enhancing the robustness of learning.
- The feedback-control mechanism efficiently encodes error signals into spike-based feedback currents, facilitating local and online synaptic weight updates compatible with event-driven neuromorphic architectures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.