Mian Wu

Mian Wu, Gavin Zhang, Sewon Min et al.

Open-ended generation tasks require outputs to satisfy diverse and often implicit task-specific evaluation rubrics. The sheer number of relevant rubrics leads to prohibitively high verification costs and incomplete assessments of a response, making reinforcement learning (RL) post-training with rubric-based rewards difficult to scale. This problem is exacerbated by the fact that often the best way to combine these rubrics into one single reward is also highly prompt-specific. We propose Reinforcement Learning with Adversarial Critic (RLAC), a post-training approach that addresses these challenges via dynamic rubric verification. Our approach employs a large language model (LLM) as a critic that dynamically identifies only the most likely failure modes (e.g., a factual error or unhandled edge case), which are then verified by an external validator to optimize both generator and critic jointly. By training both the generator and the critic, this game enhances the critic's error detection and the generator's output quality while reducing required verifications. Our experiments demonstrate that RLAC improves factual accuracy in text generation and correctness in code generation, while also outperforming exhaustive verification and reward model methods. We show that dynamic critics are more effective than fixed critics, showcasing the potential of RLAC for scaling RL post-training to free-form generation tasks.

Qi Wang, Mian Wu, Yuyang Zhang et al.

Reinforcement Learning (RL) has achieved remarkable success in various domains, yet it often relies on carefully designed programmatic reward functions to guide agent behavior. Designing such reward functions can be challenging and may not generalize well across different tasks. To address this limitation, we leverage the rich world knowledge contained in pretrained video diffusion models to provide goal-driven reward signals for RL agents without ad-hoc design of reward. Our key idea is to exploit off-the-shelf video diffusion models pretrained on large-scale video datasets as informative reward functions in terms of video-level and frame-level goals. For video-level rewards, we first finetune a pretrained video diffusion model on domain-specific datasets and then employ its video encoder to evaluate the alignment between the latent representations of agent's trajectories and the generated goal videos. To enable more fine-grained goal-achievement, we derive a frame-level goal by identifying the most relevant frame from the generated video using CLIP, which serves as the goal state. We then employ a learned forward-backward representation that represents the probability of visiting the goal state from a given state-action pair as frame-level reward, promoting more coherent and goal-driven trajectories. Experiments on various Meta-World tasks demonstrate the effectiveness of our approach.

Mian Wu, Yinling Qian, Xiangyun Liao et al.

Purpose: Segmentation of liver vessels from CT images is indispensable prior to surgical planning and aroused broad range of interests in the medical image analysis community. Due to the complex structure and low contrast background, automatic liver vessel segmentation remains particularly challenging. Most of the related researches adopt FCN, U-net, and V-net variants as a backbone. However, these methods mainly focus on capturing multi-scale local features which may produce misclassified voxels due to the convolutional operator's limited locality reception field. Methods: We propose a robust end-to-end vessel segmentation network called Inductive BIased Multi-Head Attention Vessel Net(IBIMHAV-Net) by expanding swin transformer to 3D and employing an effective combination of convolution and self-attention. In practice, we introduce the voxel-wise embedding rather than patch-wise embedding to locate precise liver vessel voxels, and adopt multi-scale convolutional operators to gain local spatial information. On the other hand, we propose the inductive biased multi-head self-attention which learns inductive biased relative positional embedding from initialized absolute position embedding. Based on this, we can gain a more reliable query and key matrix. To validate the generalization of our model, we test on samples which have different structural complexity. Results: We conducted experiments on the 3DIRCADb datasets. The average dice and sensitivity of the four tested cases were 74.8% and 77.5%, which exceed results of existing deep learning methods and improved graph cuts method. Conclusion: The proposed model IBIMHAV-Net provides an automatic, accurate 3D liver vessel segmentation with an interleaved architecture that better utilizes both global and local spatial features in CT volumes. It can be further extended for other clinical data.