Equi-Tuning: Group Equivariant Fine-Tuning of Pretrained Models
This work addresses the need for adaptable, equivariant models in various downstream tasks, offering a novel fine-tuning approach that is incremental in combining existing pretrained models with group theory.
The authors tackled the problem of making pretrained models group equivariant with minimal loss to their original representations, introducing equi-tuning as a method that achieves this across tasks like image classification, compositional generalization, and fairness in NLG, with experimental validation on benchmark datasets using models such as AlexNet, ResNet, and GPT2.
We introduce equi-tuning, a novel fine-tuning method that transforms (potentially non-equivariant) pretrained models into group equivariant models while incurring minimum $L_2$ loss between the feature representations of the pretrained and the equivariant models. Large pretrained models can be equi-tuned for different groups to satisfy the needs of various downstream tasks. Equi-tuned models benefit from both group equivariance as an inductive bias and semantic priors from pretrained models. We provide applications of equi-tuning on three different tasks: image classification, compositional generalization in language, and fairness in natural language generation (NLG). We also provide a novel group-theoretic definition for fairness in NLG. The effectiveness of this definition is shown by testing it against a standard empirical method of fairness in NLG. We provide experimental results for equi-tuning using a variety of pretrained models: Alexnet, Resnet, VGG, and Densenet for image classification; RNNs, GRUs, and LSTMs for compositional generalization; and GPT2 for fairness in NLG. We test these models on benchmark datasets across all considered tasks to show the generality and effectiveness of the proposed method.