Reference library
Generating Coherent Patterns of Activity from Chaotic Neural Networks
Paper summary
FORCE learning modifies readout/feedback or recurrent weights to organize initially chaotic rate-network activity into target patterns while feedback remains active. Rapid recursive least-squares error reduction keeps the feedback sufficiently controlled; low-amplitude targets can fail when chaos persists. The authors discuss limitations of explicit target/error signals, fast plasticity and correlation-aware updates. Initial chaos is not shown to be necessary for the learned function.
Relevance to the project
Assistant interpretation: supports studying whether feedback and learning organize internal activity rather than measuring output variance alone. Its supervision and RLS mechanism do not establish success of scorer-guided local preparation or authorize replacing the adopted learning architecture.
Targeted primary methods/discussion reading; other result sections and supplements not fully read. RLS uses information across inputs and an N-by-N correlation matrix in the basic formulation; biological locality and error sources remain unresolved. No replication or code execution.
Source record
- Reference ID
- force-learning-2009
- Archived version
- Author-hosted published Neuron63:544–557,27Aug2009 PDF
- Summary provenance
- reference-library/papers/force-learning-2009/metadata.json · summary
- Review status
- source-scoped reading; not replication
- Reading scope
- PDF pages 1,4–6,12–13 read, including main RLS equations and biological limitations. Page 5 visually checked against extracted equations; other pages/supplements not fully read.