Reproduce “Dimensions of Timescales in Neuromorphic Computing Systems” 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
Herbert Jaeger, Dirk Doorakkers, Celestine Lawrence, Giacomo Indiveri
The report surveys timescales in neuromorphic systems, distinguishing rates of change from how long information remains recoverable. It describes a prior heartbeat-classification case in which direct training on Dynap-se failed, while reservoir transfer made the task solvable with about 750 neurons. The survey is partial and offers design guidance, not a universal recipe.
Key insights
- The concept of "timescales" is highly multidimensional, involving aspects like speed vs. memory, absolute vs. relative time, and different levels of abstraction.
- More complex cognitive processing tasks fundamentally require a wider spectrum of supported memory timescales in artificial systems.
- Directly realizing a full spectrum of timescales in analog neuromorphic hardware is challenging, necessitating algorithmic and architectural workarounds like large RNNs or reservoir transfer methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.