From b4e3cbeae6cb4cf4a4b69b84a475afcd7d7e9dbe Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Thu, 26 Mar 2026 22:07:35 -0500 Subject: Add Phase 10A.6: gain requires trainable depth-aware aux, not semantic credit MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 9-branch dissection results: - zero_target crashes (-9.1%): aux must output non-zero - constant_input neutral (+0.0%): needs at least depth info - time_only works (+1.0%): h_l not needed, just depth index - shuffled/fresh_random work (+1.3-1.4%): no semantic content needed - prefit60_trainable ≈ random_trainable: prefit adds nothing - All frozen branches crash: trainability is essential Mechanism: depth-aware trainable auxiliary perturbation that diversifies block-local updates. Not semantic credit, not pure trainability. Co-Authored-By: Claude Opus 4.6 (1M context) --- NOTE.md | 20 +++++++++++++++++++- 1 file changed, 19 insertions(+), 1 deletion(-) (limited to 'NOTE.md') diff --git a/NOTE.md b/NOTE.md index 62deba6..0e8de0a 100644 --- a/NOTE.md +++ b/NOTE.md @@ -5,7 +5,7 @@ - **pilot**: Controlled iteration (commits 0b9ebb2, 7baf7ae) - **frozen**: Code at commit 0b9ebb2 for all reported results -## Status: PHASE 10A.5 — BLEND GAIN IS IMPLICIT REGULARIZATION, NOT LEARNED CREDIT +## Status: PHASE 10A.6 — GAIN REQUIRES TRAINABLE DEPTH-AWARE AUX, NOT SEMANTIC CREDIT --- @@ -587,6 +587,24 @@ Trainable Vec helps even with shuffled targets. Gaussian noise and norm scaling Phase 9A's +1.5% was not evidence of useful credit — it was an optimization dynamics effect. +### Phase 10A.6: Structured vs Semantic Auxiliary + +| Branch | final | diff | Key insight | +|--------|-------|------|-------------| +| random_trainable | 0.324 | +1.2% | works | +| shuffled_trainable | 0.325 | +1.4% | no semantics needed | +| **zero_target** | **0.221** | **-9.1%** | must output non-zero | +| fresh_random_target | 0.325 | +1.3% | stable targets not needed | +| time_only | 0.321 | +1.0% | h_l not needed, just depth | +| **constant_input** | **0.312** | **+0.0%** | needs at least depth info | +| prefit60_frozen | 0.127 | -18.4% | frozen = crash | +| prefit60_trainable | 0.321 | +1.0% | prefit ≈ random init | + +**Mechanism**: depth-aware trainable auxiliary perturbation that diversifies block-local updates. +Not semantic credit. Not pure trainability (zero_target crashes). Not state-dependent (time_only works). +Depth-awareness is the minimal requirement (constant_input fails). + ### Experiment IDs (Phase 10) - `prefit_threshold/`: Phase 10A prefit threshold curve - `blend_dissection/`: Phase 10A.5 blend mechanism dissection +- `structured_aux/`: Phase 10A.6 structured vs semantic auxiliary -- cgit v1.2.3