Author name cluster

Bo Xia

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

2 author rows

TMLR Journal 2026 Journal Article

Calibration Enhanced Decision Maker: Towards Trustworthy Sequential Decision-Making with Large Sequence Models

Haoyuan Sun
Bo Xia
Yifu Luo
Tiantian Zhang
Xueqian Wang

Offline deep reinforcement learning (offline DRL) has attracted considerable attention across various domains due to its ability to learn effective policies without direct environmental interaction. Although highly effective, the trustworthiness of agent concerns the community. Offline DRL can be categorized into three principal paradigms: model-based algorithms, model-free algorithms, and trajectory optimization. While extant research predominantly concentrates on calibration enhancement of model-based and model-free algorithms, calibration of trajectory optimization remains a rather rare topic. In this paper, we introduce the concept of Expected Agent Calibration Error (EACE), a novel metric designed to assess agent calibration. Furthermore, we rigorously prove its theoretical relationship to the state-action marginal distribution distance. Subsequently, we introduce the Calibration Enhanced Decision Maker (CEDM), which employs a binning executor to process feature distribution histograms as input for the large sequence model, thereby minimizing the state-action marginal distribution distance and enhancing the agent's calibration. A series of in-depth case studies of CEDM are carried out, with application on Decision Transformer, Decision ConvFormer, and Decision Mamba. Empirical results substantiate the robustness of EACE and demonstrate the effectiveness of CEDM in enhancing agent calibration, thereby offering valuable insights for future research on trustworthy sequential decision-making.

PDF Details

TMLR Journal 2025 Journal Article

Diffusion-RainbowPA: Improvements Integrated Preference Alignment for Diffusion-based Text-to-Image Generation

Haoyuan Sun
Bin Liang
Bo Xia
Jiaqi Wu
Yifei Zhao
Kai Qin
Yongzhe Chang
Xueqian Wang

Although rapidly increasing capabilities of text-to-image (T2I) models have profound implications across various industries, they concurrently suffer from numerous shortcomings, necessitating the implementation of effective alignment strategies with human preference. Diffusion-DPO and SPO have emerged as robust approaches for aligning diffusion-based T2I models with human preference feedback. However, they tend to suffer from text-image misalignment, aesthetic overfitting and low-quality generation. To tackle such matters, we improve the alignment paradigm through a tripartite perspective, which are the calibration enhancement (Calibration Enhanced Preference Alignment), the overfitting mitigation (Identical Preference Alignment, Jensen-Shannon Divergence Constraint) and the performance optimization (Margin Strengthened Preference Alignment, SFT-like Regularization). Furthermore, combining them with the step-aware preference alignment paradigm, we propose the Diffusion-RainbowPA, a suite of total six improvements that collectively improve the alignment performance of Diffusion-DPO. With comprehensive alignment performance evaluation and comparison, it is demonstrated that Diffusion-RainbowPA outperforms current state-of-the-art methods. We also conduct ablation studies on the introduced components that reveal incorporation of each has positively enhanced alignment performance.

PDF Details

EAAI Journal 2025 Journal Article

Dual-path aggregation transformer network for super-resolution with images occlusions and variability

Qinghui Chen
LunQian Wang
Zekai Zhang
XingHua Wang
Weilin Liu
Bo Xia
Hao Ding
Jinglin Zhang

While Transformer-based approaches have recently achieved notable success in super-resolution, their extensive computational requirements impede widespread practical adoption. High-resolution meteorological satellite cloud imagery is essential for weather analysis and forecasting. Enhancing image resolution through super-resolution techniques facilitates the accurate identification and localization of geographic features by meteorological systems. However, current super-resolution methods fail to restore the intricacies of cloud formations and complex regions fully. This research introduces a novel dual-path aggregation Transformer network (DPAT) tailored to enhance the super-resolution of meteorological satellite cloud images. The DPAT network adeptly captures cloud imagery's subtle details and textures, effectively addressing occlusions and the variability inherent in satellite imagery. It bolsters the model's ability to manage the complex attributes of cloud images through the introduction of the Dual-path Aggregation Self-Attention (DASA) mechanism and the Multi-scale Feature Aggregation Block (MFAB), thereby enhancing performance in processing intricate cloud features. The DASA mechanism synthesizes features across spatial, depth, and channel dimensions via a dual-path approach, thoroughly exploiting feature correlations. The MFAB, designed to supplant the multilayer perceptron, incorporates shift convolution and a multi-scale interaction block to augment feature information, compensating for the deficiency in local information absorption due to fixed receptive fields. Experimental outcomes indicate that DPAT delivers superior super-resolution outcomes. With a parameter count of only 32% of the Enhanced Deep Residual Network (EDSR) or 77% of the Image Restoration using Shift Window Transformer (SwinIR), DPAT matches SwinIR's performance on the satellite cloud dataset. Moreover, DPAT balances accuracy and parameter economy across various datasets. This technology is expected to improve image super-resolution capabilities in multiple fields such as human action recognition and industrial recognition, and indirectly improve the accuracy of image perception tasks.

Details DOI

ICLR Conference 2025 Conference Paper

Entropy-based Activation Function Optimization: A Method on Searching Better Activation Functions

Haoyuan Sun
Zihao Wu
Bo Xia
Pu Chang
Zibin Dong
Yifu Yuan
Yongzhe Chang
Xueqian Wang 0001

The success of artificial neural networks (ANNs) hinges greatly on the judicious selection of an activation function, introducing non-linearity into network and enabling them to model sophisticated relationships in data. However, the search of activation functions has largely relied on empirical knowledge in the past, lacking theoretical guidance, which has hindered the identification of more effective activation functions. In this work, we offer a proper solution to such issue. Firstly, we theoretically demonstrate the existence of the worst activation function with boundary conditions (WAFBC) from the perspective of information entropy. Furthermore, inspired by the Taylor expansion form of information entropy functional, we propose the Entropy-based Activation Function Optimization (EAFO) methodology. EAFO methodology presents a novel perspective for designing static activation functions in deep neural networks and the potential of dynamically optimizing activation during iterative training. Utilizing EAFO methodology, we derive a novel activation function from ReLU, known as Correction Regularized ReLU (CRReLU). Experiments conducted with vision transformer and its variants on CIFAR-10, CIFAR-100 and ImageNet-1K datasets demonstrate the superiority of CRReLU over existing corrections of ReLU. Extensive empirical studies on task of large language model (LLM) fine-tuning, CRReLU exhibits superior performance compared to GELU, suggesting its broader potential for practical applications.

Details

AAAI Conference 2025 Conference Paper

Generalizing Alignment Paradigm of Text-to-Image Generation with Preferences Through f-Divergence Minimization

Haoyuan Sun
Bo Xia
Yongzhe Chang
Xueqian Wang

Direct Preference Optimization (DPO) has recently expanded its successful application from aligning large language models (LLMs) to aligning text-to-image models with human preferences, which has generated considerable interest within the community. However, we have observed that these approaches rely solely on minimizing the reverse Kullback-Leibler divergence during alignment process between the fine-tuned model and the reference model, neglecting incorporation of other divergence constraints. In this study, we focus on extending reverse Kullback-Leibler divergence in the alignment paradigm of text-to-image models to f-divergence, which aims to garner better alignment performance as well as good generation diversity. We provide the generalized formula of text-to-image alignment paradigm under f-divergence condition and thoroughly analyze the impact of different divergence constraints on alignment process from the perspective of gradient fields. We conduct comprehensive evaluation on text-image alignment performance, human value alignment performance and generation diversity performance under different divergence constraints, and the results indicate that text-to-image alignment based on Jensen-Shannon divergence achieves the best trade-off among them. The option of divergence employed for aligning text-to-image models significantly impacts the trade-off between alignment performance (especially human value alignment) and generation diversity, which highlights the necessity of selecting an appropriate divergence for practical applications.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation

Yifu Luo
Xinhao Hu
Keyu Fan
Haoyuan Sun
Zeyu Chen
Bo Xia
Tiantian Zhang
Yongzhe Chang

Reinforcement learning (RL) has garnered increasing attention in text-to-image (T2I) generation. However, most existing RL approaches are tailored to either diffusion models or autoregressive models, overlooking an important alternative: masked generative models. In this work, we propose Mask-GRPO, the first method to incorporate Group Relative Policy Optimization (GRPO)-based RL into this overlooked paradigm. Our core insight is to redefine the transition probability, which is different from current approaches, and formulate the unmasking process as a multi-step decision-making problem. To further enhance our method, we explore several useful strategies, including removing the Kullback–Leibler constraint, applying the reduction strategy, and filtering out low-quality samples. Using Mask-GRPO, we improve a base model, Show-o, with substantial improvements on standard T2I benchmarks and preference alignment, outperforming existing state-of-the-art approaches.

PDF Details

IJCAI Conference 2022 Conference Paper

Don’t Touch What Matters: Task-Aware Lipschitz Data Augmentation for Visual Reinforcement Learning

Zhecheng Yuan
Guozheng Ma
Yao Mu
Bo Xia
Bo Yuan
Xueqian Wang
Ping Luo
Huazhe Xu

One of the key challenges in visual Reinforcement Learning (RL) is to learn policies that can generalize to unseen environments. Recently, data augmentation techniques aiming at enhancing data diversity have demonstrated proven performance in improving the generalization ability of learned policies. However, due to the sensitivity of RL training, naively applying data augmentation, which transforms each pixel in a task-agnostic manner, may suffer from instability and damage the sample efficiency, thus further exacerbating the generalization performance. At the heart of this phenomenon is the diverged action distribution and high-variance value estimation in the face of augmented images. To alleviate this issue, we propose Task-aware Lipschitz Data Augmentation (TLDA) for visual RL, which explicitly identifies the task-correlated pixels with large Lipschitz constants, and only augments the task-irrelevant pixels for stability. We verify the effectiveness of our approach on DeepMind Control suite, CARLA and DeepMind Manipulation tasks. The extensive empirical results show that TLDA improves both sample efficiency and generalization; it outperforms previous state-of-the-art methods across 3 different visual control benchmarks.

PDF Details DOI