Knowledge Composition using Task Vectors with Learned Anisotropic Scaling

TL;DR

This paper introduces a method called aTLAS that linearly combines parameter blocks of task vectors with learned anisotropic scaling, improving knowledge transfer, task composition, and generalization in pre-trained models, especially with limited data.

Contribution

It proposes a novel anisotropic scaling approach for task vector composition, exploiting low intrinsic dimensionality and enhancing transferability with less data.

Findings

Task vectors become more disentangled with anisotropic scaling.

Method improves performance in few-shot and test-time adaptation.

Reduces memory footprint and enhances flexibility in knowledge transfer.

Abstract

Pre-trained models produce strong generic representations that can be adapted via fine-tuning. The learned weight difference relative to the pre-trained model, known as a task vector, characterises the direction and stride of fine-tuning. The significance of task vectors is such that simple arithmetic operations on them can be used to combine diverse representations from different domains. This paper builds on these properties of task vectors and aims to answer (1) whether components of task vectors, particularly parameter blocks, exhibit similar characteristics, and (2) how such blocks can be used to enhance knowledge composition and transfer. To this end, we introduce aTLAS, an algorithm that linearly combines parameter blocks with different learned coefficients, resulting in anisotropic scaling at the task vector level. We show that such linear combinations explicitly exploit the low intrinsic dimensionality of pre-trained models, with only a few coefficients being the learnable parameters. Furthermore, composition of parameter blocks leverages the already learned representations, thereby reducing the dependency on large amounts of data. We demonstrate the effectiveness of our method in task arithmetic, few-shot recognition and test-time adaptation, with supervised or unsupervised objectives. In particular, we show that (1) learned anisotropic scaling allows task vectors to be more disentangled, causing less interference in composition; (2) task vector composition excels with scarce or no labeled data and is less prone to domain shift, thus leading to better generalisability; (3) mixing the most informative parameter blocks across different task vectors prior to training can reduce the memory footprint and improve the flexibility of knowledge transfer. Moreover, we show the potential of aTLAS as a PEFT method, particularly with less data, and demonstrate its scalibility.