The paper investigates the singular statistical nature of Low‑Rank Adaptation (LoRA), a popular parameter‑efficient fine‑tuning method for large neural networks. By applying the real log canonical threshold (RLCT) from singular learning theory, the authors analyze how LoRA’s learning complexity depends on its low‑rank factorization and the pretrained network. They show that the forward‑backward covariance, derived from activations and backpropagated sensitivities, governs which low‑rank perturbations are statistically observable, leading to a detailed characterization of the model’s singular geometry both locally and globally, and introduce a curvature‑corrected estimator for the RLCT.
arXiv:2610.23343v1 Announce Type: new Abstract: Low-rank adaptation (LoRA) has become the standard method for parameter-efficient fine-tuning of large pretrained models, yet theoretical explanations...
arXiv:2606. 16454v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) enables efficient adaptation of large pre-trained models to downstream tasks by parameterizing weight updates with low-rank matrices.
arXiv:2607. 27680v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has become the standard mechanism for fine-tuning large pretrained models, yet its statistical properties remain only partially understood.
Low-rank adaptation (LoRA) is the standard way to fine-tune large models, yet when its two factors are trained independently, the update ignores the geometry of the low-rank weight change it induces. We introduce LoRA-TSD, an optimizer that treats every LoRA step as a tangent vector of the fixed-rank matrix manifold and takes the spectral-norm steepest-descent step of Muon inside that tangent space, mapping the result back to the factors through a retraction native to the LoRA parametrization.
LoRA-TSD introduces a new optimizer for low‑rank adaptation (LoRA) that treats each update as a tangent vector on the fixed‑rank matrix manifold and applies a Muon‑style spectral‑norm steepest‑descent step within that tangent space. The method avoids costly full‑matrix operations and offers a retraction that is up to 2.8× cheaper than previous manifold approaches. The authors prove that their surrogate recovers LoRA‑Pro, identify the Riemannian gradient as the natural stationarity measure, and provide the first global convergence guarantees for both LoRA‑Pro and LoRA‑TSD, achieving superior performance across multiple benchmarks with Llama and Qwen models.
arXiv:2610.32849v1 Announce Type: cross Abstract: During joint training, backpropagation allows later layers to reshape earlier representations, a feedback mechanism absent from purely layer-wise lea...
arXiv:2607. 09757v1 Announce Type: cross Abstract: Low-Rank Adaptation (LoRA) has become a cornerstone of parameter-efficient fine-tuning (PEFT); however, the conventional practice of uniform rank assignment ignores the functional heterogeneity of neural layers.
arXiv:2606. 05957v1 Announce Type: new Abstract: Singular learning theory and information geometry have studied the same parameter spaces in mostly separate vocabularies: the former computes Bayesian invariants in resolved coordinates, the latter works in original coordinates under a non-degeneracy assumption that overparameterised models routinely violate.
The paper presents a comprehensive framework for comparing parameter‑efficient fine‑tuning (PEFT) methods by treating each as a smooth map from a small parameter space into weight space, with a pretrained checkpoint as the base point. It identifies three key invariants—image near the base point, fibre dynamics, and base change—that explain expressivity, training dynamics, and initialization behavior. The authors prove numerous results about common PEFT techniques (e.g., LoRA, Householder adaptation) and classify 382 methods from 138 papers using seven coordinate descriptors.
The paper introduces ISO-LoRA, an optimizer that improves rank utilization in Low‑Rank Adaptation (LoRA) by coupling factor updates through spectral descent on the induced tangent perturbation in weight space. Experiments on GPT‑2 adaptation show that standard optimizers like AdamW concentrate updates in a few singular directions, whereas ISO-LoRA distributes energy more evenly, leading to higher effective rank and better downstream performance across 0.1B‑7B models. The authors provide theoretical guarantees under a stylized spiked‑gradient model and demonstrate that ISO-LoRA consistently outperforms factor‑wise optimizers, especially at moderate‑to‑large LoRA ranks.
The paper introduces Activation Boundary Matching for Low‑Rank Adaptation (ABM‑LoRA), a task‑informed initialization strategy that uses the signs of layer‑wise pre‑activations from a brief probe adapter as targets for a fresh adapter. By training with a margin‑based hinge objective on these activation boundaries, ABM‑LoRA captures useful adaptation directions that standard LoRA initializers miss, while requiring only a few forward passes. Experiments show that ABM‑LoRA outperforms or matches existing LoRA, SVD, and gradient‑based initializers across multiple models and benchmarks, including T5‑base/GLUE, ConvNeXt‑T, Swin‑T, Qwen2.5‑1.5B, and LLaMA2‑7B.
arXiv:2607. 14018v1 Announce Type: cross Abstract: We investigate how each component of the Transformer feedforward block architecture design determines how much rank survives across depth at initialization.