From ffaa5c82b643b4313d288cdb9a4453254207f075 Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Thu, 6 Aug 2026 17:30:34 -0500 Subject: docs: require comparative random-beta baseline --- TWO_STATE_BIAS_PROGRAM.md | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-) diff --git a/TWO_STATE_BIAS_PROGRAM.md b/TWO_STATE_BIAS_PROGRAM.md index c8ed934..d84ca36 100644 --- a/TWO_STATE_BIAS_PROGRAM.md +++ b/TWO_STATE_BIAS_PROGRAM.md @@ -337,10 +337,15 @@ Stop this paper direction if: `79.15/62.75/67.10/77.75%`; innovation is `+15.00/+10.65` points over raw/constant and is `1.40` points below clean with essentially zero measured wall overhead. This is not a confirmation. Random-sign beta is not yet a - functioning comparator in the pinned positive-EP configuration: both random - sign and fixed negative beta remain at `9.9%` even with zero injected bias, - so negative-beta stability must be solved independently before comparison; - see `results/ep_bias/s3_summary.json`. + functioning comparator in the small pinned positive-EP configuration: both + random sign and fixed negative beta remain at `9.9%` even with zero injected + bias; see `results/ep_bias/s3_summary.json`. This failure is protocol-specific, + not a property of negative EP. Under the authors' original comparative-study + Conv32 architecture, relaxation counts, layer-wise rates and schedule, a + one-epoch zero-bias smoke reaches `61.40/31.55/39.10%` for positive, + negative and random-sign beta. Thus the full paper-facing EP matrix must use + the comparative-study protocol and include random-sign beta as a functioning + same-cost baseline; see `results/ep_bias/s4_summary.json`. - Coupled-learning/DCHN adapter: not implemented under this bias model. - Current score for this new paper framing: 5/10. Passing the artificial Dual Prop confirmation and the single-seed Rain S3 screen do not repair the failed -- cgit v1.2.3