论文检索

Haiying Xia, Zhongyi Huang, Yumei Tan, Shuxiang Song

Music emotion recognition is a key task in symbolic music understanding (SMER). Recent approaches have shown promising results by fine-tuning large-scale pre-trained models (e.g., MIDIBERT, a benchmark in symbolic music understanding) to map musical semantics to emotional labels. While these models effectively capture distributional musical semantics, they often overlook tonal structures, particularly musical modes, which play a critical role in emotional perception according to music psychology. In this paper, we investigate the representational capacity of MIDIBERT and identify its limitations in capturing mode-emotion associations. To address this issue, we propose a Mode-Guided Enhancement (MoGE) strategy that incorporates psychological insights on mode into the model. Specifically, we first conduct a mode augmentation analysis, which reveals that MIDIBERT fails to effectively encode emotion-mode correlations. Motivated by this observation, we further identify the MIDIBERT layer that shows the weakest emotion relevance and introduce a Mode-guided Feature-wise linear modulation injection (MoFi) framework to inject explicit mode features, thereby enhancing the model's capability in emotional representation and inference. Extensive experiments on the EMOPIA and VGMIDI datasets demonstrate that our mode injection strategy significantly improves SMER performance, achieving accuracies of 75.2% and 59.1%, respectively. These results validate the effectiveness of mode-guided modeling in symbolic music emotion recognition.

Jiaqi Wang, Liutao Yu, Xiongri Shen, Sihang Guo, Chenlin Zhou, Leilei Zhao, Yi Zhong, Zhiguo Zhang, Zhengyu Ma

Spiking neural networks (SNNs) offer a promising path toward energy-efficient speech command recognition (SCR) by leveraging their event-driven processing paradigm. However, existing SNN-based SCR methods often struggle to capture rich temporal dependencies and contextual information from speech due to limited temporal modeling and binary spike-based representations. To address these challenges, we first introduce the multi-view spiking temporal-aware self-attention (MSTASA) module, which combines effective spiking temporal-aware attention with a multi-view learning framework to model complementary temporal dependencies in speech commands. Building on MSTASA, we further propose SpikCommander, a fully spike-driven transformer architecture that integrates MSTASA with a spiking contextual refinement channel MLP (SCR-MLP) to jointly enhance temporal context modeling and channel-wise feature integration. We evaluate our method on three benchmark datasets: the Spiking Heidelberg Dataset (SHD), the Spiking Speech Commands (SSC), and the Google Speech Commands V2 (GSC). Extensive experiments demonstrate that SpikCommander consistently outperforms state-of-the-art (SOTA) SNN approaches with fewer parameters under comparable time steps, highlighting its effectiveness and efficiency for robust speech command recognition.

Ruichen Ma, Liwei Meng, Guanchao Qiao, Ning Ning, Yang Liu, Shaogang Hu

Spiking neural networks (SNNs) promise highly energy-efficient computing, but their adoption is hindered by a critical scarcity of event-stream data. This work introduces I2E, an algorithmic framework that resolves this bottleneck by converting static images into high-fidelity event streams. By simulating microsaccadic eye movements with a highly parallelized convolution, I2E achieves a conversion speed over 300x faster than prior methods, uniquely enabling on-the-fly data augmentation for SNN training. The framework's effectiveness is demonstrated on large-scale benchmarks. An SNN trained on the generated I2E-ImageNet dataset achieves a state-of-the-art accuracy of 60.50%. Critically, this work establishes a powerful sim-to-real paradigm where pre-training on synthetic I2E data and fine-tuning on the real-world CIFAR10-DVS dataset yields an unprecedented accuracy of 92.5%. This result validates that synthetic event data can serve as a high-fidelity proxy for real sensor data, bridging a long-standing gap in neuromorphic engineering. By providing a scalable solution to the data problem, I2E offers a foundational toolkit for developing high-performance neuromorphic systems. The open-source algorithm and all generated datasets are provided to accelerate research in the field.

SangEun Lee, Yubeen Lee, Eunil Park, Wonseok Chae

Understanding human emotions from images is a challenging yet essential task for vision-language models. While recent efforts have fine-tuned vision-language models to enhance emotional awareness, most approaches rely on global visual representations and fail to capture the nuanced and multi-faceted nature of emotional cues. Furthermore, most existing approaches adopt instruction tuning, which requires costly dataset construction and involves training a large number of parameters, thereby limiting their scalability and efficiency. To address these challenges, we propose MASP, a novel framework for Multi-Aspect guided emotion reasoning with Soft Prompt tuning in vision-language models. MASP explicitly separates emotion-relevant visual cues via multi-aspect cross-attention modules and guides the language model using soft prompts, enabling efficient and scalable task adaptation without modifying the base model. Our method achieves state-of-the-art performance on various emotion recognition benchmarks, demonstrating that the explicit modeling of multi-aspect emotional cues with soft prompt tuning leads to more accurate and interpretable emotion reasoning in vision-language models.

Jiayuan Huang, Runlong He, Danyal Zaman Khan, Evangelos B. Mazomenos, Danail Stoyanov, Hani Marcus, Linzhe Jiang, Matthew John Clarkson, Mobarak I. Hoque

Image-guided surgery demands adaptive, real-time decision support, yet static AI models struggle with structured task planning and providing interactive guidance. Large language models (LLMs)-powered agents offer a promising solution by enabling dynamic task planning and predictive decision support. Despite recent advances, the absence of surgical agent datasets and robust parameter-efficient fine-tuning techniques limits the development of LLM agents capable of complex intraoperative reasoning. In this paper, we introduce Surgical AI Copilot, an LLM agent for image-guided pituitary surgery, capable of conversation, planning, and task execution in response to queries involving tasks such as MRI tumor segmentation, endoscope anatomy segmentation, overlaying preoperative imaging with intraoperative views, instrument tracking, and surgical visual question answering (VQA). To enable structured agent planning, we develop the PitAgent dataset, a surgical context-aware planning dataset covering surgical tasks like workflow analysis, instrument localization, anatomical segmentation, and query-based reasoning. Additionally, we propose DEFT-GaLore, a Deterministic Energy-based Fourier Transform (DEFT) gradient projection technique for efficient low-rank adaptation of recent LLMs (e.g., LLaMA 3.2, Qwen 2.5), enabling their use as surgical agent planners. We extensively validate our agent's performance and the proposed adaptation technique against other state-of-the-art low-rank adaptation methods on agent planning and prompt generation tasks, including a zero-shot surgical VQA benchmark, demonstrating the significant potential for truly efficient and scalable surgical LLM agents in real-time operative settings.

Tongtong Feng, Xin Wang, Feilin Han, Leping Zhang, Wenwu Zhu

Swarm UAV autonomous flight for Embodied Long-Horizon (ELH) tasks is crucial for advancing the low-altitude economy. However, existing methods focus only on specific basic tasks due to dataset limitations, failing in real-world deployment for ELH tasks. ELH tasks are not mere concatenations of basic tasks, requiring handling long-term dependencies, maintaining embodied persistent states, and adapting to dynamic goal shifts. This paper presents U2UData+, the first large-scale swarm UAV autonomous flight dataset for ELH tasks and the first scalable swarm UAV data online collection and algorithm closed-loop verification platform. The dataset is captured by 15 UAVs in autonomous collaborative flights for ELH tasks, comprising 12 scenes, 720 traces, 120 hours, 600 seconds per trajectory, 4.32M LiDAR frames, and 12.96M RGB frames. This dataset also includes brightness, temperature, humidity, smoke, and airflow values covering all flight routes. The platform supports the customization of simulators, UAVs, sensors, flight algorithms, formation modes, and ELH tasks. Through a visual control window, this platform allows users to collect customized datasets through one-click deployment online and to verify algorithms by closed-loop simulation. U2UData+ also introduces an ELH task for wildlife conservation and provides comprehensive benchmarks with 9 SOTA models.

Tianwen Zhu, Guangyu Wu, Zhiwei Cao, Ruihang Wang, Jimin Jia, Yong Luo, Yonggang Wen

Existing battery State of Health (SOH) prediction approaches often struggle to provide both accurate predictions and reliable uncertainty estimates. This paper presents a novel Multi-Task Learning (MTL) framework that jointly tackles SOH prediction and provides a proxy metric for uncertainty through a unified architecture. The framework combines a Physics-Informed Neural Network (PINN) for SOH prediction with a deep autoencoding Gaussian mixture model for uncertainty modeling. Particularly, the energy score from the Gaussian mixture model serves as a proxy metric for uncertainty, where a higher score indicates potential prediction unreliability. Moreover, to enhance task-specific learning, we employ a multi-head attention mechanism that adaptively captures distinct feature relationships. Our experiments show improvements in prediction performance compared to the state-of-the-art baseline. A comprehensive evaluation on six XJTU battery benchmark datasets demonstrates that our framework achieves a prediction accuracy of 99.50% (MAPE: 0.0050) while providing reliable uncertainty quantification through the proxy metric.

Mingming Zhu, Jiahua Rao, Xiaoyu Chen, Qianmu Yuan, Yuedong Yang

Protein design is revolutionizing biotechnology, yet existing approaches struggle to balance structural foldability with functional performance. Structure-based models excel at generating stable protein backbones but often overlook critical functional properties, while protein language models capture evolutionary and functional signals but frequently predict sequences lacking structural stability. Integrating these complementary approaches remains challenging due to their inherently conflicting objectives. We present MAProt, a multi-agent framework that synergistically combines structure-based and protein language model-based methods for protein design. Each agent specializes in a distinct aspect of the design objective: the structure-based agent (e.g., ProteinMPNN) ensures compatibility with the target backbone, while protein language model-based agents (e.g., ESM, SaProt) capture evolutionary plausibility and functional potential. To reconcile conflicts and achieve optimal trade-offs, we introduce a Pareto-based negotiation module that enables effective multi-objective coordination and consensus among agents. Extensive experiments on benchmark datasets demonstrate that MAProt achieves a remarkable improvement over state-of-the-art baselines, and generalizes robustly across a range of tasks, including thermodynamic folding stability design, functional protein design, and high-affinity antibody design. These results highlight the power of collaborative optimization for advancing rational protein engineering.

Shiyan Zheng, Herun Wan, Minnan Luo, Junhang Huang

While existing social bot detectors perform well on benchmarks, their robustness across diverse real-world scenarios remains limited due to unclear ground truth and varied misleading cues. In particular, the impact of shortcut learning, where models rely on spurious correlations instead of capturing causal task-relevant features, has received limited attention. To address this gap, we conduct an in-depth study to assess how detectors are influenced by potential shortcuts based on textual features, which are most susceptible to manipulation by social bots. We design a series of shortcut scenarios by constructing spurious associations between user labels and superficial textual cues to evaluate model robustness. Results show that shifts in irrelevant feature distributions significantly degrade social bot detector performance, with an average relative accuracy drop of 32 % in the baseline models. To tackle this challenge, we propose mitigation strategies based on large language models, leveraging counterfactual data augmentation. These methods mitigate the problem from data and model perspectives across three levels, including data distribution at both the individual user text and overall dataset levels, as well as model’s ability to extract causal information. Our strategies achieve an average relative performance improvement of 56 % under shortcut scenarios.

Qinyi Zhang, Duanyu Feng, Ronghui Han, Yangshuai Wang, Hao Wang

Simulating microstructure evolution (MicroEvo) is vital for materials design but demands high numerical accuracy, efficiency, and physical fidelity. Although recent studies on deep learning (DL) offers a promising alternative to traditional solvers, the field lacks standardized benchmarks. Existing studies are flawed due to a lack of comparing specialized MicroEvo DL models with state-of-the-art spatio-temporal architectures, an overemphasis on numerical accuracy over physical fidelity, and a failure to analyze error propagation over time. To address these gaps, we introduce MicroEvoEval, the first comprehensive benchmark for image-based microstructure evolution prediction. We evaluate 14 models, encompassing both domain-specific and general-purpose architectures, across four representative MicroEvo tasks with datasets specifically structured for both short- and long-term assessment. Our multi-faceted evaluation framework goes beyond numerical accuracy and computational cost, incorporating a curated set of structure-preserving metrics to assess physical fidelity. Our extensive evaluations yield several key insights. Notably, we find that modern architectures (e.g., VMamba), not only achieve superior long-term stability and physical fidelity but also operate with an order-of-magnitude greater computational efficiency. The results highlight the necessity of holistic evaluation and identify these modern architectures as a highly promising direction for developing efficient and reliable surrogate models in data-driven materials science.

Qin-Yi Zhang, Hong Wang, Siyao Liu, Haichuan Lin, Linying Cao, Xiao-Hu Zhou, Chen Chen, Shuang-Yi Wang, Zeng-Guang Hou

Fluid–structure interaction (FSI) systems involve distinct physical domains, fluid and solid, governed by different partial differential equations and coupled at a dynamic interface. While learning-based solvers offer a promising alternative to costly numerical simulations, existing methods struggle to capture the heterogeneous dynamics of FSI within a unified framework. This challenge is further exacerbated by inconsistencies in response across domains due to interface coupling and by disparities in learning difficulty across fluid and solid regions, leading to instability during prediction. To address these challenges, we propose the Heterogeneous Graph Attention Solver (HGATSolver). HGATSolver encodes the system as a heterogeneous graph, embedding physical structure directly into the model via distinct node and edge types for fluid, solid, and interface regions. This enables specialized message-passing mechanisms tailored to each physical domain. To stabilize explicit time stepping, we introduce a novel physics-conditioned gating mechanism that serves as a learnable, adaptive relaxation factor. Furthermore, an Inter-domain Gradient-Balancing Loss dynamically balances the optimization objectives across domains based on predictive uncertainty. Extensive experiments on two constructed FSI benchmarks and a public dataset demonstrate that HGATSolver achieves state-of-the-art performance, establishing an effective framework for surrogate modeling of coupled multi-physics systems.

Jingxiao Yang, Ping He, Tianyu Du, Sun Bing, Xuhong Zhang

Recent advances in software vulnerability detection have been driven by Language Model (LM)-based approaches. However, these models remain vulnerable to adversarial attacks that exploit lexical and syntax perturbations, allowing critical flaws to evade detection. Existing black-box attacks on LM-based vulnerability detectors primarily rely on isolated perturbation strategies, limiting their ability to efficiently explore the adversarial code space for optimal perturbations. To bridge this gap, we propose HogVul, a black-box adversarial code generation framework that integrates both lexical and syntax perturbations under a unified dual-channel optimization strategy driven by Particle Swarm Optimization (PSO). By systematically coordinating two-level perturbations, HogVul effectively expands the search space for adversarial examples, enhancing the attack efficacy. Extensive experiments on four benchmark datasets demonstrate that HogVul achieves an average attack success rate improvement of 26.05% over state-of-the-art baseline methods. These findings highlight the potential of hybrid optimization strategies in exposing model vulnerabilities.

Dizhan Xue, Jing Cui, Shengsheng Qian, Chuanrui Hu, Changsheng Xu

Intelligent agents powered by large language models (LLMs) have recently demonstrated impressive capabilities and gained increasing popularity on social media platforms. While LLM agents are reshaping the ecology of social media, there exists a current gap in conducting a comprehensive evaluation of their ability to comprehend media content, understand user behaviors, and make intricate decisions. To address this challenge, we introduce SoMe, a pioneering benchmark designed to evaluate social media agents equipped with various agent tools for accessing and analyzing social media data. SoMe comprises a diverse collection of 8 social media agent tasks, 9,164,284 posts, 6,591 user profiles, and 25,686 reports from various social media platforms and external websites, with 17,869 meticulously annotated task queries. Compared with the existing datasets and benchmarks for social media tasks, SoMe is the first to provide a versatile and realistic platform for LLM-based social media agents to handle diverse social media tasks. By extensive quantitative and qualitative analysis, we provide the first overview insight into the performance of mainstream agentic LLMs in realistic social media environments and identify several limitations. Our evaluation reveals that both the current closed-source and open-source LLMs cannot handle social media agent tasks satisfactorily. SoMe provides a challenging yet meaningful testbed for future social media agents.

Ping Xu, Zaitian Wang, Zhirui Wang, Pengjiang Li, Jiajia Wang, Ran Zhang, Pengfei Wang, Yuanchun Zhou

Cell clustering is crucial for uncovering cellular heterogeneity in single-cell RNA sequencing (scRNA-seq) data by identifying cell types and marker genes. Despite its importance, existing benchmarks for scRNA-seq clustering remain fragmented, lacking standardized protocols and often omitting recent advances in artificial intelligence.To fill these gaps, we present scCluBench, a comprehensive benchmark of clustering algorithms for scRNA-seq data. scCluBench provides 36 scRNA-seq datasets collected from diverse public sources, covering multiple tissues, which are uniformly processed to ensure consistency for systematic evaluation and downstream analyses. To assess performance, we collect and reproduce a range of scRNA-seq clustering methods, including traditional, deep learning-based, graph-based, and biological foundation models. We comprehensively evaluate each method both quantitatively and qualitatively, using core performance metrics and visualization analyses. Furthermore, we construct representative downstream biological tasks, such as marker gene identification and cell type annotation, to further assess the practical utility. scCluBench then investigates the performance differences and applicability boundaries of various clustering models across diverse analytical tasks, systematically assessing their robustness and scalability in real-world scenarios. Overall, scCluBench offers a standardized and user-friendly benchmark for scRNA-seq clustering, with standardized datasets, unified evaluation protocols, and transparent analyses, facilitating informed method selection and providing valuable insights into model generalizability and application scope.

Jiancong Xie, Wentao Wei, Chi Zhang, Jiahua Rao, Yuedong Yang

Drug-drug interaction (DDI) prediction is pivotal for drug safety and clinical decision-making. Recently, subgraph-based methods utilizing knowledge graphs (KGs) and domain information have achieved promising results by extracting informative subgraphs for DDI prediction. However, existing subgraph extraction methods are typically coarse-grained and nonspecific, facing two key limitations: First, they are constrained by the vast and noisy nature of real-world KGs, making it challenging to identify the most informative substructures from the massive space of candidate subgraphs. Second, current methods often fail to exploit the molecular structural specificity of drugs to selectively extract relevant subgraphs, lacking effective integration of molecular structure information with knowledge graph context. To address these challenges, we propose RISE-DDI, a novel framework for Reinforced-based Informative Subgraph Extraction approach for drug-drug interaction prediction. Specifically, RISE-DDI formulates the subgraph extraction as a Markov Decision Process (MDP) and leverages a deep reinforcement learning (RL) agent to dynamically and adaptively extract the most informative and context-specific subgraphs for each drug pair. The agent is guided by a learnable structure-aware reward model that considers both the topological context from the knowledge graph and the molecular features of the drug pairs, thereby encouraging the selection of subgraphs that are both structurally relevant and biologically informative. Extensive experiments on DDI benchmark datasets demonstrate that our method outperforms state-of-the-art baselines in both transductive and inductive scenarios, achieving improvements of up to 20%. Furthermore, visualization analyses of the extracted subgraphs highlight the interpretability of our model, providing insights into the underlying mechanisms of drug interactions.

Xiaohua Wu, Lin Li, Kaize Shi, Xiaohui Tao, Jianwei Zhang, Yuefeng Li

The response behaviors observed in online user-generated content (UGC) frequently demonstrate non-linear characteristics, such as conditional branching and selective avoidance. These patterns present additional challenges for ensuring the trustworthiness of Large Language Model (LLMs) reasoning, particularly as their unidirectional, left-to-right inference mechanisms may not adequately capture such complex reasoning dynamics. To address this, we propose a Forest of Thought Explanation (FoTE), a novel prompting that models the selective avoidance in UGC while ensuring explanation consensus through reasoning paths across all decision sub-trees. FoTE firstly generates various reasoning paths through an adaptive CoT prompting. Each generated thought is subsequently evaluated through cooperative game theory to quantify its fair influence. The thoughts with the top-k contribution scores are preserved and randomly sampled to emulate selective avoidance for the next reasoning iteration. Through extensive evaluations across three open-source LLMs and two established social science problems (spanning four benchmark datasets), FoTE demonstrates superior success rates compared to competing prompting strategies. Notably, its performance gains increase with the strength of selective avoidance in social problems. The trustworthiness of our FoTE is enhanced by the incorporation of (1) a solid theoretical foundation and (2) a transparent reasoning path that converges toward consensus.

Qianyang Wu, Jingwei Lv, Zilong Zhang, Feifei Cui

Predicting drug–target interactions (DTIs) is a fundamental task in computational drug discovery, yet it remains challenging under distribution shifts and limited training data. Existing approaches often suffer from poor generalization, weak cross-modal alignment between molecular and protein representations, and vulnerability to noisy supervision.We propose ESP-DTI, a unified framework designed to enhance generalization by integrating large-scale protein language models with curriculum learning and cross-modal contrastive alignment. Specifically, we leverage ESM-2 to encode context-aware protein representations and adopt a CLIP-style contrastive objective to align drug and protein embeddings in a shared latent space. To further improve learning robustness, we introduce a progressive curriculum sampling strategy that dynamically schedules training instances based on model confidence, enabling a gradual shift from easy to hard examples.Experimental results on four benchmark datasets demonstrate that ESP-DTI consistently outperforms state-of-the-art baselines, achieving a +3.1% improvement in average accuracy. Ablation studies confirm the complementary benefits of each component, validating their collective contribution to robust and generalizable DTI prediction.Our work underscores the effectiveness of combining pretrained protein language models with structured training curricula and cross-modal contrastive learning for reliable DTI prediction under real-world, distribution-shifted conditions.

Mengmeng Wei, Lei Wang, Zhu-Hong You, Pengwei Hu, Bowei Zhao, Zhi-An Huang, Yu-An Huang, Haicheng Yi

CircRNA-miRNA interaction (CMI) plays a pivotal role in disease therapeutics and drug discovery. However, existing methods face several challenges in modeling complex biological networks and zero-shot learning scenarios. Biological networks encapsulate rich biological information, yet current approaches often fail to fully exploit this depth. Moreover, zero-shot prediction requires models to identify new interactions without relying on previously observed samples, imposing stringent requirements on generalization capabilities. To address these limitations, we propose a dual-channel learning framework leveraging State space modeling for Zero-shot CMI prediction (ZeroStem). ZeroStem first enhances the biological relevance of node using prior knowledge, and employs a graph Transformer to extract macro-topological representations. Subsequently, it generates semantic subgraphs based on meta-paths to focus on specific biological relationships, utilizing the Mamba to extract micro-semantic representations via state space modeling. Finally, macro-topological and micro-semantic representations are seamlessly integrated through linear transformation and residual connections, enabling high-precision zero-shot CMI prediction. Extensive experiments on multiple benchmark datasets demonstrate that ZeroStem significantly outperforms existing methods, validating its efficiency and robust generalization in CMI prediction. Case studies further illustrate that ZeroStem offers novel insights into the molecular mechanisms underlying intricate disease-associated networks.

Ying Wang, Guoheng Huang, Xueyuan Gong, Xinxin Wang, Xiaochen Yuan

Automated auscultation advances the detection of respiratory diseases, especially in areas with limited resources where traditional diagnostic methods are unavailable. On the other hand, the scarcity of auscultation datasets limits the automation performance, prompting the needs for data augmentation methods. However, most of the existing methods neglect the difference in acoustic sounds that requires personalized augmentation strategies. To address this, we propose a Progressive-Adaptive Spectral Augmentation (PASA), which is one of the first paradigms to adaptively select the best augmentation strategy for each sample. The PASA innovatively treats augmentation selection problem as a Markov Decision Process (MDP), creating an alternating loop between the diagnostic model and the augmentation selection. The agent selects the optimal augmentation operations and magnitudes via a task-specific design, including state construction, action sampling, Hybrid Batch-Sample (HBS) strategy execution, and reward guidance. The HBS strategy initially applies uniform augmentation across mini-batches while collecting sample-specific performance statistics. When model performance stabilizes, it transits to sample-level augmentation based on accumulated difficulty assessments. This two-phase design balances computational complexity with personalization. Extensive experiments across three benchmark datasets demonstrate that the PASA outperforms the state-of-the-art methods, pioneering a transformative paradigm for adaptive data augmentation in automated auscultation.

Bing Wang, Ximing Li, Yanjun Wang, Changchun Li, Lin Yuanbo Wu, Buyu Wang, Shengsheng Wang

Multimodal Misinformation Detection (MMD) refers to the task of detecting social media posts involving misinformation, where the post often contains text and image modalities. However, by observing the MMD posts, we hold that the text modality may be much more informative than the image modality because the text generally describes the whole event/story of the current post but the image often presents partial scenes only. Our preliminary empirical results indicate that the image modality exactly contributes less to MMD. Upon this idea, we propose a new MMD method named RETSIMD. Specifically, we suppose that each text can be divided into several segments, and each text segment describes a partial scene that can be presented by an image. Accordingly, we split the text into a sequence of segments, and feed these segments into a pre-trained text-to-image generator to augment a sequence of images. We further incorporate two auxiliary objectives concerning text-image and image-label mutual information, and further post-train the generator over an auxiliary text-to-image generation benchmark dataset. Additionally, we propose a graph structure by defining three heuristic relationships between images, and use a graph neural network to generate the fused features. Extensive empirical results validate the effectiveness of RETSIMD.