MixPHM: Redundancy-Aware Parameter-Efficient Tuning for Low-Resource Visual Question Answering
This addresses the challenge of adapting large models efficiently for low-resource VQA tasks, representing a strong specific gain rather than a broad paradigm shift.
The paper tackles the problem of computationally expensive and storage-inefficient full finetuning of Vision-Language Models for low-resource Visual Question Answering by proposing MixPHM, a redundancy-aware parameter-efficient tuning method that outperforms full finetuning on datasets like VQA v2, GQA, and OK-VQA.
Recently, finetuning pretrained Vision-Language Models (VLMs) has been a prevailing paradigm for achieving state-of-the-art performance in Visual Question Answering (VQA). However, as VLMs scale, finetuning full model parameters for a given task in low-resource settings becomes computationally expensive, storage inefficient, and prone to overfitting. Current parameter-efficient tuning methods dramatically reduce the number of tunable parameters, but there still exists a significant performance gap with full finetuning. In this paper, we propose MixPHM, a redundancy-aware parameter-efficient tuning method that outperforms full finetuning in low-resource VQA. Specifically, MixPHM is a lightweight module implemented by multiple PHM-experts in a mixture-of-experts manner. To reduce parameter redundancy, MixPHM reparameterizes expert weights in a low-rank subspace and shares part of the weights inside and across experts. Moreover, based on a quantitative redundancy analysis for adapters, we propose Redundancy Regularization to reduce task-irrelevant redundancy while promoting task-relevant correlation in MixPHM representations. Experiments conducted on VQA v2, GQA, and OK-VQA demonstrate that MixPHM outperforms state-of-the-art parameter-efficient methods and is the only one consistently surpassing full finetuning.