Learning Perturbations to Extrapolate Your LLM

• URL: http://arxiv.org/abs/2605.13284v1
• Date: Wed, 13 May 2026 10:00:58 GMT
• Title: Learning Perturbations to Extrapolate Your LLM
• Abstract summary: We propose a framework where token prefixes are perturbed by a learnable transformation of a continuous latent vector within an embedding space.<n>To overcome the challenge of an intractable marginal likelihood, we derive estimating equations for model parameters and optimize them via gradient descent.
• Score: 12.34345212532974
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent advancements in large language models demonstrate that injecting perturbations can substantially enhance extrapolation performance. However, current approaches often rely on discrete perturbations with fixed designs, which limits their flexibility. In this work, we propose a framework where token prefixes are perturbed by a learnable transformation of a continuous latent vector within an embedding space. To overcome the challenge of an intractable marginal likelihood, we derive unbiased estimating equations for model parameters and optimize them via stochastic gradient descent. We establish the statistical properties of the resulting estimator in over-parameterized regimes. Empirical evaluations on both synthetic and real-world datasets demonstrate that our proposal yields significant gains in out-of-domain settings over a range of state-of-the-art baseline methods.

Related papers

• Particle-based Generalised Stochastic Optimisation [0.0]
  We develop a class of diffusion-based particle optimisation methods for loss functions with intractable gradients.<n>We introduce mean-field dynamics and its interacting-particle approximations, which contain several existing algorithms as special cases.<n>We show that the continuous-time particle system admits a non-asymptotic error bound.
  arXiv  Detail & Related papers  (2026-08-03T19:59:47Z)
• Joint Model and Data Sparsification via the Marginal Likelihood [53.29070892356214]
  We propose jointly learning individual feature and sample relevancies, enabling simultaneous model and data sparsification.<n>This symmetric pruning of model and data offers a natural extension that preserves conjugacy.<n> Empirical results across diverse regression tasks affirm that a joint ARD approach consistently yields both sparse and robust prediction models.
  arXiv  Detail & Related papers  (2026-05-28T13:26:53Z)
• Non-Parametric Rehearsal Learning via Conditional Mean Embeddings [88.89267783967263]
  We propose a non-parametric rehearsal learning approach for textitavoiding undesired future (AUF) problems.<n>Specifically, we use kernel machinery to reformulate the AUF objective into a unified representation that disentangles desirability modeling from action-induced distributional changes.
  arXiv  Detail & Related papers  (2026-05-09T15:30:52Z)
• Revisiting Zeroth-Order Optimization: Minimum-Variance Two-Point Estimators and Directionally Aligned Perturbations [57.179679246370114]
  We identify the distribution of random perturbations that minimizes the estimator's variance as the perturbation stepsize tends to zero.<n>Our findings reveal that such desired perturbations can align directionally with the true gradient, instead of maintaining a fixed length.
  arXiv  Detail & Related papers  (2025-10-22T19:06:39Z)
• On the Optimal Construction of Unbiased Gradient Estimators for Zeroth-Order Optimization [57.179679246370114]
  A potential limitation of existing methods is the bias inherent in most perturbation estimators unless a stepsize is proposed.<n>We propose a novel family of unbiased gradient scaling estimators that eliminate bias while maintaining favorable construction.
  arXiv  Detail & Related papers  (2025-10-22T18:25:43Z)
• Nonparametric Data Attribution for Diffusion Models [57.820618036556084]
  Data attribution for generative models seeks to quantify the influence of individual training examples on model outputs.<n>We propose a nonparametric attribution method that operates entirely on data, measuring influence via patch-level similarity between generated and training images.
  arXiv  Detail & Related papers  (2025-10-16T03:37:16Z)
• Efficient Group Lasso Regularized Rank Regression with Data-Driven Parameter Determination [2.847099287022546]
  High-dimensional regression often suffers from heavy-tailed noise and outliers, which can severely undermine the reliability of least-squares based methods.<n>To improve robustness, we adopt a non-smooth Wilcoxon score based rank objective and incorporate structured group sparsity regularization.<n>We also introduce a data-driven, simulation-based tuning rule and further establish a finite-sample error bound for the resulting estimator.
  arXiv  Detail & Related papers  (2025-10-13T15:45:58Z)
• Bayesian Nonparametrics Meets Data-Driven Distributionally Robust Optimization [29.24821214671497]
  Training machine learning and statistical models often involve optimizing a data-driven risk criterion. We propose a novel robust criterion by combining insights from Bayesian nonparametric (i.e., Dirichlet process) theory and a recent decision-theoretic model of smooth ambiguity-averse preferences. For practical implementation, we propose and study tractable approximations of the criterion based on well-known Dirichlet process representations.
  arXiv  Detail & Related papers  (2024-01-28T21:19:15Z)
• Parameter Estimation in DAGs from Incomplete Data via Optimal Transport [24.740382124473975]
  We develop a theoretical framework and support it with extensive empirical evidence demonstrating the robustness and versatility of our approach. We show that not only can our method effectively recover the ground-truth parameters but it also performs comparably or better than competing baselines on downstream applications.
  arXiv  Detail & Related papers  (2023-05-25T10:54:36Z)
• Improving Generalization via Uncertainty Driven Perturbations [107.45752065285821]
  We consider uncertainty-driven perturbations of the training data points. Unlike loss-driven perturbations, uncertainty-guided perturbations do not cross the decision boundary. We show that UDP is guaranteed to achieve the robustness margin decision on linear models.
  arXiv  Detail & Related papers  (2022-02-11T16:22:08Z)
• MINIMALIST: Mutual INformatIon Maximization for Amortized Likelihood Inference from Sampled Trajectories [61.3299263929289]
  Simulation-based inference enables learning the parameters of a model even when its likelihood cannot be computed in practice. One class of methods uses data simulated with different parameters to infer an amortized estimator for the likelihood-to-evidence ratio. We show that this approach can be formulated in terms of mutual information between model parameters and simulated data.
  arXiv  Detail & Related papers  (2021-06-03T12:59:16Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.