From a6d9cc53f77906305dcefa454d77151c1996405f Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Thu, 6 Aug 2026 16:08:32 -0500 Subject: docs: position SDIL as debiasing plug-in --- TWO_STATE_BIAS_PROGRAM.md | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'TWO_STATE_BIAS_PROGRAM.md') diff --git a/TWO_STATE_BIAS_PROGRAM.md b/TWO_STATE_BIAS_PROGRAM.md index 58bdf6c..b0df1b2 100644 --- a/TWO_STATE_BIAS_PROGRAM.md +++ b/TWO_STATE_BIAS_PROGRAM.md @@ -18,6 +18,10 @@ one feedback backbone. Its central hypothesis is: > neutral, per-cell prediction removes the identifiable component before it is > consolidated into synapses. +SDIL is presented as a local debiasing plug-in for existing two-state learners, +not as a new optimizer on an accuracy--cost Pareto frontier. Comparisons use +matched budgets where possible and report unmatched overhead in a table. + “Two-state” includes equilibrium propagation (one recurrent network at free and nudged equilibria), coupled learning, contrastive Hebbian learning, Dual Propagation, and related positive/negative-state rules. It does not imply that @@ -212,8 +216,9 @@ same bias variable and neutral observation have an exact, auditable meaning. Every backbone reports its native clean endpoint, biased raw endpoint, best bias-specific baseline, SDIL and oracle. “Beat” means a paired advantage under bias at matched architecture/data/training, together with no meaningful clean -regression and an explicit cost coordinate. It does not mean that SDIL must -beat BP or centered EP on clean accuracy. +regression, a matched observation protocol and disclosed implementation +overhead. It does not mean that SDIL must beat BP or centered EP on clean +accuracy. | family | natural/independent bias | strongest required baseline | scale axis | |---|---|---|---| -- cgit v1.2.3