Yi-Lin Sung

Task-Circuit Quantization: Leveraging Knowledge Localization and Interpretability for Compression

Glider: Global and Local Instruction-Driven Expert Router

Merge, Then Compress: Demystify Efficient SMoE with Hints from Its Routing Policy

SELMA: Learning and Merging Skill-Specific Text-to-Image Experts with Auto-Generated Data

ECoFLaP: Efficient Coarse-to-Fine Layer-Wise Pruning for Vision-Language Models

Dynamic Adapter Merging for Continual Video Question-Answering Learning