Chuanrui Hu, Xingze Gao, Zuyi Zhou, Dannong Xu, Yi Bai, Xintong Li, Hui Zhang, Tong Li, Chong Zhang, Lidong Bing, Yafeng Deng. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Juwei Yue, Chuanrui Hu, Jiawei Sheng, Zuyi Zhou, Wenyuan Zhang, Tingwen Liu, Li Guo, Yafeng Deng. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Vehicular networks draw much attention as the essential communication system of vehicles, especially with the development of autonomous driving. The channel resource is shared and also contented by each user, therefore, quality of service (QoS) is hard to be guaranteed. Existing solutions, especially machine learning based algorithms [1], cannot fully address dynamic neighbor's status because the feature size varies according to the varying number of neighbors. In this work, we make a practical dataset for resource allocation purpose using ns-3 and SUMO, which has been mainly used for vehicle-to-vehicle communication. Various features are collected. Furthermore, a graph convolutional network (GCN) [2] is used to perform the classification of transmission status, success or failure. The prediction accuracy is improved by 15% than that of LSTM thanks to the graph representation of data, which implies an alleviation of packets collision.

Vehicular networks have stringent quality of service (QoS) requirements in terms of reliability, throughput, and latency. With the emergence of diverse services for autonomous driving, the resource contention in vehicle-to-vehicle communication can cause unavoidable packet loss and, therefore, must be handled for safety. Moreover, different levels of QoS should be defined for each service and task; however, existing solutions can neither provide a resource allocation scheme for any level of QoS requirement nor serve new stringent services without configuration or modification. We propose adistributed hierarchical deep Q-network(DH-DQN) to handle resource contention specifically. Thus, an intelligence resource management (I-RM) scheme is designed to serve on-demand QoSs. We first formulate the problem to address multiple QoS requirements, which extends the coverage of resource management tasks for on-demand stringent services. From the perspective of transmission pattern, we designed a hierarchical DQN structure that deals with resource block contention in a fully distributed manner and a state-action framework that enables a numerically defined service demand. In addition, a target$\epsilon$-greedy is proposed to accelerate convergence, and a modified transfer learning algorithm is used to enhance learning performance for various levels of service. Through extensive simulations, we demonstrated that the proposed DH-DQN can learn successful transmission patterns to meet different levels of multiple QoS requirements.

Masked image modeling (MIM) has shown great promise for self-supervised learning (SSL) yet been criticized for learning inefficiency. We believe the insufficient utilization of training signals should be responsible. To alleviate this issue, we introduce a conceptually simple yet learning-efficient MIM training scheme, termedDisjointMasking withJointDistillation (DMJD). For disjoint masking (DM), we sequentially sample multiple masked views per image in a mini-batch with the disjoint regulation to raise the usage of tokens for reconstruction in each image while keeping the masking rate of each view. For joint distillation (JD), we adopt a dual branch architecture to respectively predict invisible (masked) and visible (unmasked) tokens with superior learning targets. Rooting in orthogonal perspectives for training efficiency improvement, DM and JD cooperatively accelerate the training convergence yet not sacrificing the model generalization ability. Concretely, DM can train ViT with less effective training epochs (at most$3.7\times$less time-consuming) to report competitive performance. With JD, our DMJD clearly improves the linear probing classification accuracy, up to 3.4$\%$. On fine-grained downstream tasks like semantic segmentation, object detection,etc., our DMJD also presents superior generalization compared with state-of-the-art SSL methods.

To extend the network's life cycle in wireless sensor networks, clustering plays an important role in balancing energy consumption. In this paper, we propose a novel clustering method based on reinforcement learning that integrates cluster head selection and cluster formation as one step. It considers both energy efficiency and inter cluster interference in the model-free design, thus achieving longer network lifetime and higher quality of packet transmission. To the best of our knowledge, our work is the first paper that integrates cluster head selection and cluster formation using reinforcement learning. Our extensive simulation results show that the proposed method improves the network lifetime by 65% and 29% compared with Low Energy Adaptive Clustering Hierarchy (LEACH) and Greedy Energy Efficient Clustering Scheme (GEECS), respectively, while the data transmission success rate is also increased by 42% and 31%, respectively.

Optimizing molecules with desired properties is a crucial step in de novo drug design. While translation-based methods have achieved initial success, they continue to face the challenge of the “exposure bias” problem. The challenge of preventing the “exposure bias” problem of molecule optimization lies in the need for both positive and negative molecules of contrastive learning. That is because generating positive molecules through data augmentation requires domain-specific knowledge, and randomly sampled negative molecules are easily distinguished from the real molecules. Hence, in this work, we propose a molecule optimization method called GPMO, which leverages a gradient perturbation-based contrastive learning method to prevent the “exposure bias” problem in translation-based molecule optimization. With the assistance of positive and negative molecules, GPMO is able to effectively handle both real and artificial molecules. GPMO is a molecule optimization method that is conditioned on matched molecule pairs for drug discovery. Our empirical studies show that GPMO outperforms the state-of-the- art molecule optimization methods. Furthermore, the negative and positive perturbations improve the robustness of GPMO.

Vision-language pre-training (VLP) on large-scale datasets has shown premier performance on various downstream tasks. In contrast to plenty of available benchmarks with English corpus, large-scale pre-training datasets and downstream datasets with Chinese corpus remain largely unexplored. In this work, we build a large-scale high-quality Chinese Cross-Modal Benchmark named CCMB for the research community, which contains the currently largest public pre-training dataset Zero and five human-annotated fine-tuning datasets for downstream tasks. Zero contains 250 million images paired with 750 million text descriptions, plus two of the five fine-tuning datasets are also currently the largest ones for Chinese cross-modal downstream tasks. Along with the CCMB, we also develop a VLP framework named R2D2, applying a pre-Ranking + Ranking strategy to learn powerful vision-language representations and a two-way distillation method (i.e., target-guided Distillation and feature-guided Distillation) to further enhance the learning capability. With the Zero and the R2D2 VLP framework, we achieve state-of-the-art performance on twelve downstream datasets from five broad categories of tasks including image-text retrieval, image-text matching, image caption, text-to-image generation, and zero-shot image classification. The datasets, models, and codes are available at https://github.com/yuxie11/R2D2

Identification of potential targets for known bioactive compounds and novel synthetic analogs is of considerable significance. In silico target fishing (TF) has become an alternative strategy because of the expensive and laborious wet-lab experiments, explosive growth of bioactivity data and rapid development of high-throughput technologies. However, these TF methods are based on different algorithms, molecular representations and training datasets, which may lead to different results when predicting the same query molecules. This can be confusing for practitioners in practical applications. Therefore, this study systematically evaluated nine popular ligand-based TF methods based on target and ligand-target pair statistical strategies, which will help practitioners make choices among multiple TF methods. The evaluation results showed that SwissTargetPrediction was the best method to produce the most reliable predictions while enriching more targets. High-recall similarity ensemble approach (SEA) was able to find real targets for more compounds compared with other TF methods. Therefore, SwissTargetPrediction and SEA can be considered as primary selection methods in future studies. In addition, the results showed that k = 5 was the optimal number of experimental candidate targets. Finally, a novel ensemble TF method based on consensus voting is proposed to improve the prediction performance. The precision of the ensemble TF method outperforms the individual TF method, indicating that the ensemble TF method can more effectively identify real targets within a given top-k threshold. The results of this study can be used as a reference to guide practitioners in selecting the most effective methods in computational drug discovery.

Machine learning-based scoring functions (MLSFs) have become a very favorable alternative to classical scoring functions because of their potential superior screening performance. However, the information of negative data used to construct MLSFs was rarely reported in the literature, and meanwhile the putative inactive molecules recorded in existing databases usually have obvious bias from active molecules. Here we proposed an easy-to-use method named AMLSF that combines active learning using negative molecular selection strategies with MLSF, which can iteratively improve the quality of inactive sets and thus reduce the false positive rate of virtual screening. We chose energy auxiliary terms learning as the MLSF and validated our method on eight targets in the diverse subset of DUD-E. For each target, we screened the IterBioScreen database by AMLSF and compared the screening results with those of the four control models. The results illustrate that the number of active molecules in the top 1000 molecules identified by AMLSF was significantly higher than those identified by the control models. In addition, the free energy calculation results for the top 10 molecules screened out by the AMLSF, null model and control models based on DUD-E also proved that more active molecules can be identified, and the false positive rate can be reduced by AMLSF.

The rational design of chemical entities with desired properties for a specific target is a long-standing challenge in drug design. Generative neural networks have emerged as a powerful approach to sample novel molecules with specific properties, termed as inverse drug design. However, generating molecules with biological activity against certain targets and predefined drug properties still remains challenging. Here, we propose a conditional molecular generation net (CMGN), the backbone of which is a bidirectional and autoregressive transformer. CMGN applies large-scale pretraining for molecular understanding and navigates the chemical space for specified targets by fine-tuning with corresponding datasets. Additionally, fragments and properties were trained to recover molecules to learn the structure-properties relationships. Our model crisscrosses the chemical space for specific targets and properties that control fragment-growth processes. Case studies demonstrated the advantages and utility of our model in fragment-to-lead processes and multi-objective lead optimization. The results presented in this paper illustrate that CMGN has the potential to accelerate the drug discovery process.

Cracking the entangling code of protein-ligand interaction (PLI) is of great importance to structure-based drug design and discovery. Different physical and biochemical representations can be used to describe PLI such as energy terms and interaction fingerprints, which can be analyzed by machine learning (ML) algorithms to create ML-based scoring functions (MLSFs). Here, we propose the ML-based PLI capturer (ML-PLIC), a web platform that automatically characterizes PLI and generates MLSFs to identify the potential binders of a specific protein target through virtual screening (VS). ML-PLIC comprises five modules, including Docking for ligand docking, Descriptors for PLI generation, Modeling for MLSF training, Screening for VS and Pipeline for the integration of the aforementioned functions. We validated the MLSFs constructed by ML-PLIC in three benchmark datasets (Directory of Useful Decoys-Enhanced, Active as Decoys and TocoDecoy), demonstrating accuracy outperforming traditional docking tools and competitive performance to the deep learning-based SF, and provided a case study of the Serine/threonine-protein kinase WEE1 in which MLSFs were developed by using the ML-based VS pipeline in ML-PLIC. Underpinning the latest version of ML-PLIC is a powerful platform that incorporates physical and biological knowledge about PLI, leveraging PLI characterization and MLSF generation into the design of structure-based VS pipeline. The ML-PLIC web platform is now freely available at http://cadd.zju.edu.cn/plic/.