Jie Ouyang, Tingyue Pan, Mingyue Cheng, Ruiran Yan, Yucong Luo, Jiaying Lin, Qi Liu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The emergence of Large Language Models (LLMs) has unified language generation tasks and revolutionized human-machine interaction. However, in the realm of image generation, a unified model capable of handling various tasks within a single framework remains largely unexplored. In this work, we introduce OmniGen, a new diffusion model for unified image generation. OmniGen is characterized by the following features: 1) Unification: OmniGen not only demonstrates text-to-image generation capabilities but also inherently supports various downstream tasks, such as image editing, subject-driven generation, and visual-conditional generation. 2) Simplicity: The architecture of OmniGen is highly simplified, eliminating the need for additional plugins. Moreover, compared to existing diffusion models, it is more user-friendly and can complete complex tasks end-to-end through instructions without the need for extra intermediate steps, greatly simplifying the image generation workflow. 3) Knowledge Transfer: Benefit from learning in a unified format, OmniGen effectively transfers knowledge across different tasks, manages unseen tasks and domains, and exhibits novel capabilities. We also explore the model’s reasoning capabilities and potential applications of the chain-of-thought mechanism. This work represents the first attempt at a general-purpose image generation model, and we will release our resources at https://github.com/VectorSpaceLab/OmniGen to foster future advancements.

Multi-task learning (MTL) has become increasingly prevalent in e-commerce recommender systems. However, existing MTL methods, particularly those utilizing the Multi-gate Mixture-of-Experts (MMoE) architecture, face challenges due to their implicit routing mechanisms. These mechanisms can inadvertently lead to negative knowledge transfer, failing to resolve conflicts among tasks and resulting in gradient contradictions on shared parameters. Such issues undermine the generalization capability of MTL models across various tasks. To address these limitations, we introduce the Task Information Decoupling Model (TIDM), designed to alleviate negative transfer by decoupling task knowledge. TIDM incorporates two innovative modules following the expert layer: the Maximize Information Aggregation Module (MIA) and the Automatic Information Selection Module (AIS). The MIA module employs an auxiliary loss to filter out irrelevant task information and aggregates task-specific knowledge using a dissimilar self-attention network. Subsequently, the AIS module automatically selects the most pertinent task-specific information to facilitate task tower learning. Our experiments demonstrate that TIDM outperforms five contemporary MTL models across two datasets, showcasing its effectiveness in extracting task-specific information. This advancement is crucial for enhancing the performance of recommender systems in e-commerce and other complex domains.