Reference library
Towards deep learning with segregated dendrites
Paper summary
A computational network uses electrotonically segregated feedforward and feedback compartments and differences between ordinary and target-driven phases to obtain local plateau-related teaching signals. Fixed random feedback, phase coordination and settling/averaging remain assumptions. Connections between local hidden objectives and output improvement have explicit conditions, including a small output discrepancy and a Jacobian condition; extra depth did not provide unrestricted improvement.
Relevance to the project
Shows why separation of instruction from forward activity and explicit timing matter in a local learning proposal. Our known scorer/context differs from a biologically generated target. The study motivates recording differentiation and update ordering; it adopts no compartment, phase or updater change.
Biologically inspired computational proposal rather than direct animal evidence. Plateau interpretation, phase control and apical-to-basal plasticity implementation are unresolved. Formal proofs, all figures/supplements and peer-review correspondence were not fully audited. Separate acquired arXiv v3 PDF was identity checked but not the read version.
Source record
- Reference ID
- doi-10-7554-elife.22901
- Archived version
- Published eLife article 6:e22901 (2017), archived full-text XML via Europe PMC.; arXiv:1610.00161v3, 7 April 2017; version printed on original first page; not the read source.
- Summary provenance
- reference-library/papers/doi-10-7554-elife.22901/metadata.json · summary
- Review status
- Targeted source-grounded review; no experimental reproduction or exhaustive proof/figure audit
- Reading scope
- Published XML original acquired. Read Introduction, selected Results on phases/plateaus/local objectives, Discussion, selected Methods and narrative/theorem conditions. No complete formal derivation, figure, supplement or code reproduction.