Xin Geng

Longhua Li, Lei Qi, Xin Geng

Training and deploying multiple vision transformer (ViT) models for different resource constraints is costly and inefficient. To address this, we propose transforming a pre-trained ViT into a stratified knowledge-density super-network, where knowledge is hierarchically organized across weights. This enables flexible extraction of sub-networks that retain maximal knowledge for varying model sizes. We introduce \textbf{W}eighted \textbf{P}CA for \textbf{A}ttention \textbf{C}ontraction (WPAC), which concentrates knowledge into a compact set of critical weights. WPAC applies token-wise weighted principal component analysis to intermediate features and injects the resulting transformation and inverse matrices into adjacent layers, preserving the original network function while enhancing knowledge compactness. To further promote stratified knowledge organization, we propose \textbf{P}rogressive \textbf{I}mportance-\textbf{A}ware \textbf{D}ropout (PIAD). PIAD progressively evaluates the importance of weight groups, updates an importance-aware dropout list, and trains the super-network under this dropout regime to promote knowledge stratification. Experiments demonstrate that WPAC outperforms existing pruning criteria in knowledge concentration, and the combination with PIAD offers a strong alternative to state-of-the-art model compression and model expansion methods.

Shunxin Guo, Jiaqi Lv, Xin Geng

We introduce Ring-topology Decentralized Federated Learning (RDFL) for distributed model training, aiming to avoid the inherent risks of centralized failure in server-based FL. However, RDFL faces the challenge of low information-sharing efficiency due to the point-to-point communication manner when handling inherent data heterogeneity. Existing studies to mitigate data heterogeneity focus on personalized optimization of models, ignoring that the lack of shared information constraints can lead to large differences among models, weakening the benefits of collaborative learning. To tackle these challenges, we propose a Divide-and-conquer RDFL framework (DRDFL) that uses a feature generation model to extract personalized information and invariant shared knowledge from the underlying data distribution, ensuring both effective personalization and strong generalization. Specifically, we design a \textit{PersonaNet} module that encourages class-specific feature representations to follow a Gaussian mixture distribution, facilitating the learning of discriminative latent representations tailored to local data distributions. Meanwhile, the \textit{Learngene} module is introduced to encapsulate shared knowledge through an adversarial classifier to align latent representations and extract globally invariant information. Extensive experiments demonstrate that DRDFL outperforms state-of-the-art methods in various data heterogeneity settings.

Miaosen Zhang, Yixuan Wei, Zhen Xing et al.

Modern vision models are trained on very large noisy datasets. While these models acquire strong capabilities, they may not follow the user's intent to output the desired results in certain aspects, e.g., visual aesthetic, preferred style, and responsibility. In this paper, we target the realm of visual aesthetics and aim to align vision models with human aesthetic standards in a retrieval system. Advanced retrieval systems usually adopt a cascade of aesthetic models as re-rankers or filters, which are limited to low-level features like saturation and perform poorly when stylistic, cultural or knowledge contexts are involved. We find that utilizing the reasoning ability of large language models (LLMs) to rephrase the search query and extend the aesthetic expectations can make up for this shortcoming. Based on the above findings, we propose a preference-based reinforcement learning method that fine-tunes the vision models to distill the knowledge from both LLMs reasoning and the aesthetic models to better align the vision models with human aesthetics. Meanwhile, with rare benchmarks designed for evaluating retrieval systems, we leverage large multi-modality model (LMM) to evaluate the aesthetic performance with their strong abilities. As aesthetic assessment is one of the most subjective tasks, to validate the robustness of LMM, we further propose a novel dataset named HPIR to benchmark the alignment with human aesthetics. Experiments demonstrate that our method significantly enhances the aesthetic behaviors of the vision models, under several metrics. We believe the proposed algorithm can be a general practice for aligning vision models with human values.