Mixture-of-Experts (MoE) models route each token to a subset of expert networks, increasing capacity while keeping per-token computation sparse. Many deployed MoEs use a fixed number of active experts across layers and tasks, despite evidence that layer roles, expert redundancy, and required capacity vary with depth and input difficulty. Existing layer-wise methods usually determine allocations offline and reuse them across tasks, while token-level methods rely on local routing signals without task-level context. We propose MetaNet, a support-set controller that predicts an expert-retention threshold and a bounded routing bias for each layer while keeping the backbone, experts, and original router frozen. On DeepSeek-MoE-16B-Chat, MetaNet provides a tunable accuracy–expert-activation trade-off. Relative to fixed k=6, a conservative setting activates 3.61 experts on average (40% fewer) with comparable MMLU accuracy (0.489 versus 0.474); an aggressive setting activates 2.28 experts on average (62% fewer) at 0.438 accuracy. Trained on MMLU, MetaNet transfers to C-Eval without retraining, activating 2.90 experts on average (52% fewer than fixed k=6) at 0.386 accuracy.

Referring Expression Comprehension (REC), which aims to ground a local visual region via natural language, is a task that heavily relies on multimodal alignment. Most existing methods utilize powerful pre-trained models to transfer visual/linguistic knowledge by full fine-tuning. However, full fine-tuning the entire backbone not only breaks the rich prior knowledge embedded in the pre-training, but also incurs significant computational costs. Motivated by the recent emergence of Parameter-Efficient Transfer Learning (PETL) methods, we aim to solve the REC task in an effective and efficient manner. Directly applying these PETL methods to the REC task is inappropriate, as they lack the specific-domain abilities for precise local visual perception and visual-language alignment. Therefore, we propose a novel framework of Multimodal Prior-guided Parameter Efficient Tuning, namely MaPPER. Specifically, MaPPER comprises Dynamic Prior Adapters guided by a aligned prior, and Local Convolution Adapters to extract precise local semantics for better visual perception. Moreover, the Prior-Guided Text module is proposed to further utilize the prior for facilitating the cross-modal alignment. Experimental results on three widely-used benchmarks demonstrate that MaPPER achieves the best accuracy compared to the full fine-tuning and other PETL methods with only 1.41% tunable backbone parameters.