When Edit Localization Amplifies Relative Selection Bias: Gradient Geometry, Target Mismatch, and Importance Weighting

Shengwei Zhang; Haoda Dai; Yifei Li; Yuheng Song

cs.CLSep 12, 2026

When Edit Localization Amplifies Relative Selection Bias: Gradient Geometry, Target Mismatch, and Importance Weighting

Authors: Shengwei Zhang, Haoda Dai, Yifei Li, Yuheng Song

Organizations: University of Pennsylvania

Abstract

Human corrections identify editable spans, but the examples receiving corrections may come from a selective feedback channel. We analyze this interaction at a fixed model checkpoint by decomposing a localized gradient into edited and retained untouched components. Squared relative selection bias is a ratio of quadratics whose derivative has the sign of an explicit quadratic polynomial. Localization can increase, decrease, or nonmonotonically change this diagnostic; its direction depends on component biases and geometry. Oracle importance weighting recovers the population mean for each fixed localization objective, but these objectives have different targets. Against one common full-gradient target, we derive the finite-sample mean-squared error, an analytic optimal retention coefficient, and a fixed-clipping extension. Exact finite-population calculations and 10,000 Monte Carlo repetitions per sample size verify the identities and counterexamples. Public human-post-edit experiments use two translation directions and pretrained models, with declared synthetic selection. An English-German extension differentiates 73.89 million native parameters. In all three declared settings, hard localization has higher relative bias but lower absolute bias than full retention. Untouched-component biases are nonzero and selected component means have negative inner products, so the general criterion applies where the simple unbiased/aligned explanation fails. Output-bias diagnostics show the same endpoint ordering of relative bias across both directions, with one interior maximum. Mechanisms reuse each language's records and include a mixture; they are not independent replications. The evidence separates relative amplification from absolute gradient error and establishes estimation properties, without inferring translation-quality gains or identifying actual complaint propensities.

Figures & tables

Figure 1: Exact finite-population geometry. Smaller

\alpha

means stronger localization. A: unbiased untouched gradients and aligned means yield relative-bias amplification as

\alpha

decreases, although absolute bias stays fixed. B: selection in the untouched component reverses the effect. C: opposing means produce an interior maximum. Values are exact population quantities; no sampling intervals are needed.

Figure 2: Exact MSE to one full traffic-gradient target. Left: the oracle IPW optimum moves toward full retention as sample size increases; dots show the analytic minimizers. Right: joint retention and population-normalized clipping frontier at

n=4

. The cap grid contains 100 log-spaced values, and retention is optimized analytically at each cap. The marked solution is the best on that cap grid, not a claim of a universal deployment setting.

$n$	Estimator	MSE to $\mu_{1}$	Monte Carlo SE
16	Unweighted	0.7842	0.0050
16	Oracle IPW	0.5487	0.0082
16	Self-normalized	0.5480	0.0050
16	Population-clipped ( $M=2$ )	0.3987	0.0039
256	Unweighted	0.6375	0.0012
256	Oracle IPW	0.0339	0.0004

Table 1: Executed Monte Carlo results for full retention (

\alpha=1

), using 10,000 repetitions per sample size. The clipped estimator uses a fixed population normalizer and cap

M=2

, while the self-normalized estimator has a random sample normalizer. Lower MSE is better for this declared gradient target.

Figure 3: Finite-sample error to the same full-gradient target. Left: hard localization remains mismatched even as oracle IPW estimates its own localized target accurately. Right: full-retention oracle IPW and self-normalization approach the full target, while fixed clipping retains asymptotic bias. Error bars show approximate 95% Monte Carlo intervals for mean MSE.

Direction	Selection feature	$\rho(0)$	$\rho(1)$	$\lVert\varepsilon\rVert$
en-de	Edit fraction	0.454	0.347	0.0031
en-de	Source length	0.210	0.131	0.0014
en-de	Both	0.227	0.154	0.0016
en-zh	Edit fraction	0.220	0.088	0.0044
en-zh	Source length	0.093	0.037	0.0021
en-zh	Both	0.115	0.053	0.0022

Table 2: Shared output-bias gradients under synthetic selection, with 256 reporting records per direction. The ratio uses the selected mean norm as in Equation ( 4 );

\varepsilon=b_{U}

is the untouched-component bias. Values are exact for the declared finite populations, up to numerical precision.

Figure 4: Measured output-bias geometry on public human post-edits, under synthetic selection. Lines evaluate the exact ratio of quadratics from the stored population means; grid points are separately enumerated. English–German edit-based selection has a small interior maximum at

\alpha\approx 0.026984

, retained in the analysis despite the larger endpoint difference.

Selection	$\rho_{Q}(0)$	$\rho_{Q}(1)$	$\lVert b_{0}\rVert$	$\lVert b_{1}\rVert$	$\lVert\varepsilon\rVert$
Edit	0.44748	0.33102	0.49340	0.65602	0.47136
Length	0.19267	0.10497	0.12684	0.20144	0.16385
Edit + length	0.22455	0.14286	0.18330	0.27333	0.21668

Table 3: Completed native-parameter gradient diagnostics on the fixed public reporting population. The primary relative measure uses the selected mean in its denominator. Hard localization has larger relative bias but smaller absolute bias than full retention in all three settings. These are finite-population gradient measurements under declared synthetic selection, not translation-quality changes.

Figure 5: Native-parameter relative and absolute gradient bias. Curves are evaluated from the saved six moments; 21 grid points and the exact derivative coefficients are exported. The two panels expose the distinction between a normalized discrepancy and its numerator. No parameter update is performed.

Explore similar work

From Spectral Saliency to Finite Deletion: Evaluating Selective Language Model Edits
Oct 13, 2026Ibne Farabi Shihab, Joyanta Jyoti MondalLanguage Model Editing LLM Pruning
Department of Computer and Information Sciences, University of Delaware, USA
Exact Finite Attention Responses From RoPE Derivatives
Sep 12, 2026Julie Huang, Maggie Chlon, Gregory Gutin +1Causal Interventions in Language Models Softmax Attention
Hassana Labs · Department of Computer Science, Royal Holloway, University of London · University of Oxford
Form Over Content In Gradient-Based Data Attribution Methods
Sep 17, 2026Sunwoo Kim, Seokwon Jung, Sohyung Kim +2Gradient-Based Attribution Instruction-Tuning Data Selection
KAIST

Oct 13, 2026J/K move · Enter open · S save

From Spectral Saliency to Finite Deletion: Evaluating Selective Language Model Edits

Ibne Farabi Shihab, Joyanta Jyoti Mondal

Department of Computer and Information Sciences, University of Delaware, USA

Language Model Editing LLM Pruning

Open paper AskPDF arXiv

Deleting singular components offers a precise way to remove a capability from a language model, yet the first-order spectral saliency used to rank components predicts only a local loss change, whereas deletion is a finite intervention. In this paper, we study how prediction error affects selective editing, where selective editing requires degradation of a target task while preserving others. We decompose the error into derivative transfer across item splits and nonlinear response along the deletion path. We establish that outer interval screening retains every feasible candidate when every candidate's intervals contain its finite validation loss changes on every task. We evaluate these effects using one answer-only negative log-likelihood objective for search, validation, and test, with disjoint calibration and evaluation masks. The primary configuration targets arithmetic in Mistral-7B-Instruct-v0.3 while preserving code and question answering; extensions cover Llama-3.1-8B-Instruct and reversed target roles. In the primary study, which deletes 1024 coordinates from a support of 7168, calibrated intervals cover 968 of 1024 evaluation masks. Outer screening reaches the direct-selected edit after 25 verification checks, whereas direct verification abstains at a 32-check cap, but it rejects 4 of 74 feasible masks. All three selected edits pass validation, yet only one meets every test constraint. Interval coverage, feasible-candidate retention, and independent test preservation are therefore distinct properties and require separate assessment.

Abstract

Explore similar work

From Spectral Saliency to Finite Deletion: Evaluating Selective Language Model Edits

Exact Finite Attention Responses From RoPE Derivatives

Form Over Content In Gradient-Based Data Attribution Methods