Reproduce “Encoding Event-Based Data With a Hybrid SNN Guided Variational Auto-encoder in Neuromorphic Hardware” 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
Kenneth Stewart, Andreea Danielescu, Timothy Shea, Emre Neftci
Researchers developed a Hybrid Guided Variational Autoencoder, integrating a Spiking Neural Network (SNN) encoder, to create disentangled latent representations from event-based sensor data. This method achieved 87% classification accuracy on the DVSGesture dataset and enabled similarity learning for novel, unlabeled gestures, with a proof-of-concept implementation on Intel Loihi neuromorphic hardware.
Key insights
- The guided VAE framework is critical for creating disentangled and interpretable latent representations of event-based data, significantly improving classification accuracy and feature separation.
- Spiking Neural Network (SNN) encoders effectively process asynchronous event streams, preserving spatio-temporal information more robustly than conventional CNN encoders for this data type.
- The learned latent space facilitates generalization, allowing the model to accurately cluster and pseudo-label novel, unlabeled gestures based on their similarity to known classes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.