Published in SemEval-2025, 2025
This paper presents our approach to SemEval-2025 Task 4, which focuses on unlearning in semantic understanding. We propose a model merging strategy that consolidates alternative model versions to enforce effective knowledge removal.
Recommended citation: Xu, H., et al. (2025). "ZJUKLAB at SemEval-2025 Task 4: Unlearning via Model Merging." SemEval-2025.
Download Paper
Published in arXiv preprint, 2025
This paper introduces Relearn, a novel framework for machine unlearning in large language models that enables effective removal of specific knowledge while preserving model performance through a learning-based approach.
Recommended citation: Xu, H., et al. (2025). "Relearn: Unlearning via Learning for Large Language Models." arXiv preprint arXiv:2408.15168.
Download Paper
Published in arXiv preprint, 2024
This paper introduces a dynamic correction decoding strategy for multimodal large language models (MLLMs) that leverages visual information to detect and correct hallucinations during text generation, significantly reducing factual errors.
Recommended citation: Xu, H., et al. (2024). "MLLM Can See? Dynamic Correction Decoding for Hallucination Mitigation." arXiv preprint arXiv:2406.00000.
Download Paper