cs.CLAug 3, 2026

Not the Dimension, the Norm: What Matters in Gradient-Free Weight Perturbation of Language Models

Authors: Taeyeong Kim, Ahhyun Kim, TaeHyeon Kim, Unggi Lee

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Adapting a language model to a task no longer requires training all of its weights, and a line of parameter-efficient methods has driven the trainable count from billions down to a handful of scalars. Gradient-free adaptation, which samples random weight perturbations and keeps the ones that score well, has not followed that trajectory and still perturbs every entry of the weight tensor. It is unknown whether that full-weight search is necessary, and more fundamentally which property of a perturbation makes it work at all, because existing methods vary the search space, the perturbation scale, and the aggregation together. We resolve this by intervening on one factor at a time inside a fixed pipeline, holding candidate scoring and voting constant while we vary the search dimension, the subspace that carries the perturbation, and its norm. Perturbing a frozen frame of 12 to 16 scalars stays 1.8 accuracy points behind full-weight search on average across 49 model-benchmark cells, trailing it in 36 of them. Neither the dimension nor the choice of basis explains that performance. A random frame whose Grassmann overlap with the SVD frame is at chance level performs identically once a single scale factor is matched, and at large scales the SVD directions collapse first. What survives is the perturbation norm, whose usable range closes within a factor of five across seven models and stays flat inside. The perturbation norm is therefore the one factor with a failure mode, and its safe region transfers across scale and family. The design question narrows from which subspace to perturb to how hard to shake.

Explore similar work

CardsList
Mar 25, 2026cs.CL

Perturbation: A simple and efficient adversarial tracer for representation learning in language models

Linguistic representation learning in deep neural language models (LMs) has been studied for decades, but finding representations in LMs remains an unsolved problem. On the one hand, unconstrained alignments may trivialize the notion of representation (Sutter et al., 2025); on the other, even recently popularized linear approaches may not always be faithful to natural model behavior (Arora et al. 2024). Here we escape this dilemma by reconceptualizing representations not as patterns of activation but as conduits for learning. Our approach is simple: we perturb an LM by fine-tuning it on a single adversarial example and measure how this perturbation "infects" other examples. Perturbation makes no geometric assumptions, and unlike other methods, it does not find representations where it should not (e.g., in untrained LMs). But in trained LMs, perturbation reveals structured transfer at multiple linguistic grain sizes, suggesting that LMs both generalize along representational lines and acquire linguistic abstractions from experience alone.
Joshua Rozner, Cory Shain
PerturbationsModel ActivationsRepresentation LearningAdversarial ExamplesOnline Learning
AskPDF
Sep 22, 2026cs.LG

Modular Norm RandOpt: Population-Efficient Ensembling through Architecture-Aware Perturbations

RandOpt samples weight-perturbed language models and ensembles top-ranked candidates through plurality voting, but its global perturbation scale ignores heterogeneous module geometry. We propose Modular Norm RandOpt, an architecture-aware sampling method using module-wise natural norms and calibrated scales while preserving selection and voting. It outperforms RandOpt using 3×3\times fewer candidates on Countdown and at least 12×12\times fewer on GSM8K, with corresponding wall-clock savings. Evaluations across seven tasks and three Qwen scales (0.50.5B--33B) show higher mean accuracy than RandOpt on Countdown, GSM8K, and MATH-500 at every scale. The gains extend to Llama 3.2 33B and Gemma 3 44B on Countdown and GSM8K. On Qwen2.5-1.5B, our ensembles also achieve higher mean accuracy than iterative baselines on both tasks at comparable main-run evaluation budgets. On GSM8K, a tail-density diagnostic implies only a 1.21.2--1.8×1.8\times candidate reduction, while most ensemble improvement is associated with more favorable correct-expert support. These results highlight perturbation geometry as a key design choice for population-efficient, gradient-free search around pretrained models.
Kirato Yoshihara, Hiroaki Hamade
Topic pending
AskPDF
Jun 30, 2026cs.LG

Fora: From Weight-Space to Function-Space Protection in Capability-Preserving Fine-Tuning

Full fine-tuning adapts large language models to new tasks but can erode capabilities they already possess. Existing remedies protect through proxies such as parameter distances, importance penalties, output matching, or dominant singular directions of the weights, but none directly asks which activation directions the preserved capability relies on. We argue that a capability is characterized more faithfully by the activation subspace it induces than by the singular geometry of the weight matrix, and develop function-space protection, instantiated as FORA (Function-space Orthogonal Residual Adaptation). From label-free calibration inputs, FORA estimates, per layer, the principal directions QQ of the input-activation covariance and forms a right projector PQ=I−QQTP_Q = I - QQ^T. Paired with a left projector PUP_U from the weight SVD, the update is ΔW=PUMPQ+U2DδV2TΔW = P_U M P_Q + U_2 D_δ V_2^T: a high-capacity branch structurally barred from reading capability-relevant function directions, plus a narrow spectral channel for controlled plasticity. The construction extends to parameter-efficient adaptation via M→(α/r)BAM \to (α/r) BA. Across three settings on Qwen3-1.7B, including COGS and GSM8K learned while preserving translation and translation learned while preserving math, FORA consistently improves preservation over weight-space projection and standard regularization, with only a small new-task trade-off in the math-preservation setting. A controlled ablation isolating the projection source shows that the advantage comes not from projection itself, but from projecting onto capability-derived rather than weight-derived directions. Code is available at https://github.com/zrui239/FORA.
Rui Zhou, Tianci Xie
Parameter-Efficient AdaptationWeight-SpaceCovariance Estimation
AskPDF