Yujie Feng

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.

Xuanhua Yin, Dingxin Zhang, Yu Feng et al.

Existing rotation-invariant point cloud masked autoencoders (MAE) rely on random masking strategies that overlook geometric structure and semantic coherence. Random masking treats patches independently, failing to capture spatial relationships consistent across orientations and overlooking semantic object parts that maintain identity regardless of rotation. We propose a dual-stream masking approach combining 3D Spatial Grid Masking and Progressive Semantic Masking to address these fundamental limitations. Grid masking creates structured patterns through coordinate sorting to capture geometric relationships that persist across different orientations, while semantic masking uses attention-driven clustering to discover semantically meaningful parts and maintain their coherence during masking. These complementary streams are orchestrated via curriculum learning with dynamic weighting, progressing from geometric understanding to semantic discovery. Designed as plug-and-play components, our strategies integrate into existing rotation-invariant frameworks without architectural changes, ensuring broad compatibility across different approaches. Comprehensive experiments on ModelNet40, ScanObjectNN, and OmniObject3D demonstrate consistent improvements across various rotation scenarios, showing substantial performance gains over the baseline rotation-invariant methods.