Zexin Lu

Guangyuan Shi, Zexin Lu, Xiaoyu Dong et al.

Aligning large language models (LLMs) through supervised fine-tuning is essential for tailoring them to specific applications. Recent studies suggest that alignment primarily adjusts a model's presentation style rather than its foundational knowledge, indicating that only certain components of the model are significantly impacted. To uncover how alignment affects model behavior at a granular level, we propose identifying which layers within LLMs are most critical to the alignment process. Our approach, named ILA, involves learning a binary mask for the parameter changes in each layer during alignment, as an indicator of layer significance. Experimental results reveal that, despite substantial differences in alignment datasets, the important layers of a model identified by ILA exhibit nearly 90\% overlap, highlighting fundamental patterns in LLM alignment. The results also indicate that freezing non-essential layers improves overall model performance, while selectively tuning the most critical layers significantly enhances fine-tuning efficiency with minimal performance loss. Finally, we discuss how these findings extend from LLM alignment to reasoning.

Yujie Feng, Xujia Wang, Zexin Lu et al.

Continual learning (CL) is crucial for deploying large language models (LLMs) in dynamic real-world environments without costly retraining. While recent model ensemble and model merging methods guided by parameter importance have gained popularity, they often struggle to balance knowledge transfer and forgetting, mainly due to the reliance on static importance estimates during sequential training. In this paper, we present Recurrent-KIF, a novel CL framework for Recurrent Knowledge Identification and Fusion, which enables dynamic estimation of parameter importance distributions to enhance knowledge transfer. Inspired by human continual learning, Recurrent-KIF employs an inner loop that rapidly adapts to new tasks while identifying important parameters, coupled with an outer loop that globally manages the fusion of new and historical knowledge through redundant knowledge pruning and key knowledge merging. These inner-outer loops iteratively perform multiple rounds of fusion, allowing Recurrent-KIF to leverage intermediate training information and adaptively adjust fusion strategies based on evolving importance distributions. Extensive experiments on two CL benchmarks with various model sizes (from 770M to 13B) demonstrate that Recurrent-KIF effectively mitigates catastrophic forgetting and enhances knowledge transfer.

Yujie Feng, Liming Zhan, Zexin Lu et al.

Regular updates are essential for maintaining up-to-date knowledge in large language models (LLMs). Consequently, various model editing methods have been developed to update specific knowledge within LLMs. However, training-based approaches often struggle to effectively incorporate new knowledge while preserving unrelated general knowledge. To address this challenge, we propose a novel framework called Geometric Knowledge Editing (GeoEdit). GeoEdit utilizes the geometric relationships of parameter updates from fine-tuning to differentiate between neurons associated with new knowledge updates and those related to general knowledge perturbations. By employing a direction-aware knowledge identification method, we avoid updating neurons with directions approximately orthogonal to existing knowledge, thus preserving the model's generalization ability. For the remaining neurons, we integrate both old and new knowledge for aligned directions and apply a "forget-then-learn" editing strategy for opposite directions. Additionally, we introduce an importance-guided task vector fusion technique that filters out redundant information and provides adaptive neuron-level weighting, further enhancing model editing performance. Extensive experiments on two publicly available datasets demonstrate the superiority of GeoEdit over existing state-of-the-art methods.

Tao Wang, Wenjun Xia, Zexin Lu et al.

Due to the presence of metallic implants, the imaging quality of computed tomography (CT) would be heavily degraded. With the rapid development of deep learning, several network models have been proposed for metal artifact reduction (MAR). Since the dual-domain MAR methods can leverage the hybrid information from both sinogram and image domains, they have significantly improved the performance compared to single-domain methods. However,current dual-domain methods usually operate on both domains in a specific order, which implicitly imposes a certain priority prior into MAR and may ignore the latent information interaction between both domains. To address this problem, in this paper, we propose a novel interactive dualdomain parallel network for CT MAR, dubbed as IDOLNet. Different from existing dual-domain methods, the proposed IDOL-Net is composed of two modules. The disentanglement module is utilized to generate high-quality prior sinogram and image as the complementary inputs. The follow-up refinement module consists of two parallel and interactive branches that simultaneously operate on image and sinogram domain, fully exploiting the latent information interaction between both domains. The simulated and clinical results demonstrate that the proposed IDOL-Net outperforms several state-of-the-art models in both qualitative and quantitative aspects.