paper-with-me

Papers

Task Matrices: Linear Maps for Cross-Model Finetuning Transfer

2025-12-16 · Darrin O' Brien, Dhikshith Gajulapalli, Eric Xia arxiv

Results in interpretability suggest that large vision and language models learn implicit linear encodings when models are biased by in-context prompting. However, the existence of similar linear representations in more general adaptation regimes has not yet been demonstrated. In this work, we develop the concept of a task matrix, a linear transformation from a base to finetuned embedding state. We demonstrate that for vision and text models and ten different datasets, a base model augmented with a task matrix achieves results surpassing linear probes, sometimes approaching finetuned levels. Our results validate the existence of cross-layer linear encodings between pretrained and finetuned architectures. Moreover, we show that a data-based approximation for such encodings is both efficient and generalizable to multiple domains. We make our implementation publicly available.

📄 PDF Abstract BibTeX arXiv:2512.14880

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

On the Emergence of Cross-Task Linearity in the Pretraining-Finetuning Paradigm

2024-02-06 · Zhanpeng Zhou, Zijun Chen, Yilan Chen, Bo Zhang 외

The pretraining-finetuning paradigm has become the prevailing trend in modern deep learning. In this work, we discover an intriguing linear phenomenon in models that are initialized from a common pretrained checkpoint an…

Orthogonal Model Merging

2026-02-05 · Sihan Yang, Kexuan Shi, Weiyang Liu arxiv

Merging finetuned Large Language Models (LLMs) has become increasingly important for integrating diverse capabilities into a single unified model. However, prevailing model merging methods rely on linear arithmetic in Eu…

Kaleidoscope: An Efficient, Learnable Representation For All Structured Linear Maps

2020-12-29 · ICLR 2020 1 · Tri Dao, Nimit S. Sohoni, Albert Gu, Matthew Eichhorn 외

Modern neural network architectures use structured linear transformations, such as low-rank matrices, sparse matrices, permutations, and the Fourier transform, to improve inference speed and reduce memory usage compared …

Allimage-classificationImage Classificationspeech-recognition+1

From GaLore to WeLore: How Low-Rank Weights Non-uniformly Emerge from Low-Rank Gradients

2024-07-15 · Ajay Jaiswal, Lu Yin, Zhenyu Zhang, Shiwei Liu 외

Modern Large Language Models (LLMs) are composed of matrices with billions of elements, making their storage and processing quite demanding in terms of computational resources and memory usage. Being significantly large,…

GPU

Frobenius-Type Norms and Inner Products of Matrices and Linear Maps with Applications to Neural Network Training

2023-11-26 · Roland Herzog, Frederik Köhne, Leonie Kreis, Anton Schiela

The Frobenius norm is a frequent choice of norm for matrices. In particular, the underlying Frobenius inner product is typically used to evaluate the gradient of an objective with respect to matrix variable, such as thos…