diff options
| -rw-r--r-- | RAIN_EP_DILLAVOU_MATRIX.md | 15 | ||||
| -rw-r--r-- | RAIN_EP_RELEASED_PROFILE.md | 15 |
2 files changed, 17 insertions, 13 deletions
diff --git a/RAIN_EP_DILLAVOU_MATRIX.md b/RAIN_EP_DILLAVOU_MATRIX.md index 094d9d6..3fe39e2 100644 --- a/RAIN_EP_DILLAVOU_MATRIX.md +++ b/RAIN_EP_DILLAVOU_MATRIX.md @@ -2,8 +2,11 @@ ## Question -Can a local neutral predictor remove a fixed hardware update offset without -using a larger EP nudging voltage? +The fixed-offset experiment is only an implementation sanity check. With an +exact neutral observation, subtracting a strictly constant offset is already +an oracle and cannot demonstrate an SDIL advantage. The scientific question +is whether SDIL improves downstream accuracy when local hardware bias changes +with device state and a stored constant is therefore insufficient. The primary corruption is the model in Dillavou et al., Eq. (7). Every local parameter update receives a fixed, unknown, parameter-specific offset after @@ -20,9 +23,11 @@ state-drift ratio is zero in all exact-model experiments. The correction is local and uses no backpropagation. With the teaching input disabled, the update circuit exposes `B_i`. The predictor stores this neutral -measurement and subtracts it from subsequent updates. Under the strictly -constant model, constant calibration and SDIL innovation are expected to -coincide. A difference between them is neither predicted nor claimed. +measurement and subtracts it from subsequent updates. Under the strictly +constant model, constant calibration, SDIL innovation, and oracle subtraction +are equivalent after the exact neutral observation. A difference between them +is neither predicted nor claimed, and recovery in this setting is not evidence +for SDIL. ## Frozen author protocol diff --git a/RAIN_EP_RELEASED_PROFILE.md b/RAIN_EP_RELEASED_PROFILE.md index eaefe87..eb6a16f 100644 --- a/RAIN_EP_RELEASED_PROFILE.md +++ b/RAIN_EP_RELEASED_PROFILE.md @@ -97,14 +97,13 @@ the author's comparative ConvHopfieldEnergy32 network. Holdout accuracy was: | released state-dependent profile | affine SDIL, cadence 10 | 42.35% | | released state-dependent profile | exact oracle subtraction | 46.00% | -This is positive evidence that a local neutral measurement can rescue a -centered-EP learner from a damaging constant hardware offset: both local -correctors recover 45.45% from 10.00%, close to the 46.00% clean endpoint. -It is negative evidence for the stronger state-dependent claim at the frozen -R1 setting: affine SDIL is 1.50 percentage points below the matched local -intercept and 3.65 points below the oracle. The state-dependent result must -therefore be treated as unconfirmed until a paired multi-seed accuracy study -shows otherwise. No residual-bias diagnostic can override this endpoint. +The constant-offset rows are an implementation sanity check only. Because the +neutral observation is exact, measuring a strictly constant offset once and +subtracting it is already oracle correction; the matching intercept and SDIL +rows contain no scientific advantage. The only substantive R1 comparison is +the state-dependent profile. There, affine SDIL is 1.50 percentage points +below the matched intercept tracker and 3.65 points below exact oracle +subtraction. No residual-bias diagnostic can override this accuracy result. ## R2 paired-seed accuracy check |