Reproduce “Distributed Representations Enable Robust Multi-Timescale Symbolic Computation 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
Madison Cotteret, Hugh Greatorex, Alpha Renner, Junren Chen, Emre Neftci, Huaqiang Wu, Giacomo Indiveri, Martin Ziegler and 1 more
This research introduces a method to embed multi-timescale symbolic computation into recurrent spiking neural networks using high-dimensional distributed representations. The single-shot weight learning scheme enables robust and scalable implementation of deterministic finite automata on neuromorphic hardware, including memristive crossbars and Intel Loihi 2, even with significant synaptic nonidealities.
Key insights
- High-dimensional distributed representations (Vector Symbolic Architectures) intrinsically provide robustness against imprecise component functioning and hardware nonidealities, making them well-suited for reliable neuromorphic computation.
- Complex multi-timescale dynamics of deterministic finite automata can be embedded into RSNNs through a single-shot, non-iterative weight construction using outer products, bypassing the challenges of gradient-based learning.
- The strategic combination of autoassociative terms for state stability and input-masked transition terms, leveraging hypervector pseudo-orthogonality, allows for elegant and robust state transitions without requiring delicate parameter balancing or external modulation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.