What explains the success of cross-modal fine-tuning with ORCA?
This work provides incremental insights for researchers in cross-modal learning by clarifying the roles of ORCA's components, potentially guiding more efficient fine-tuning strategies.
The paper investigates the contributions of embedder training and model fine-tuning in ORCA's cross-modal fine-tuning success, finding that embedder training is unnecessary for 2D tasks and only moderately beneficial for 1D tasks, with model fine-tuning being the key factor in most cases.
ORCA (Shen et al., 2023) is a recent technique for cross-modal fine-tuning, i.e., applying pre-trained transformer models to modalities beyond their training data. The technique consists primarily of training an embedder and fine-tuning the embedder and model. Despite its high performance on a variety of downstream tasks, we do not understand precisely how each of these components contribute to ORCA's success. Therefore, we run a series of ablations and find that embedder training does not help 2D tasks at all, contrary to what the original paper posits. In 1D tasks, some amount of embedder training is necessary but more is not better. In 4 out of 6 datasets we experiment with, it is model fine-tuning that makes the biggest difference. Through our ablations and baselines, we contribute a better understanding of the individual components of ORCA.