Reference library
Optimal sequence memory in driven random networks
Paper summary
Dynamic mean-field analysis of driven random leaky-tanh networks distinguishes local instability from asymptotic conditional chaos. Replicas share the same weights and fluctuating input. A locally expansive but nonchaotic regime appears, and the network contribution to linear sequential memory peaks there. Stronger chaos can obscure input information. Input-driven activity and correlations persist even when initial-state differences decay.
Relevance to the project
Assistant interpretation: compare sensitivity under the same experience with retained-input performance instead of maximizing autonomous variability. Useful memory before conditional chaos is a plausible comparison, not a prediction already established for plastic F/H networks.
Large-network fixed random connectivity, independent Gaussian white inputs, continuous time and a specific linear reconstruction task/readout approximation. The normalized capacity bound is task-specific. Appendices and final journal changes were not audited; no new experiment was run.
Source record
- Reference ID
- driven-network-memory-2018
- Archived version
- arXiv:1603.01880v3, submitted22Sep2017; manuscript dated25Sep2017; related2018 journal work, not final typeset version
- Summary provenance
- reference-library/papers/driven-network-memory-2018/metadata.json · summary
- Review status
- source-scoped reading; not replication
- Reading scope
- Main text PDF pages 1–10 read; page 8 figures/captions visually verified from rendered original. Appendices 11–18 not fully read or proof-audited.