Tong Chen, JiaWei Guo, Yuxi Li, Baiming Chen, Houxing Ren, Zhang Zhiwei, Yunxiang Zhang, Hanyang Xia, Kun Liang, Zhaoran Fan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Cerebrovascular diseases (CVDs) such as aneurysms, arteriovenous malformations, stenosis, and Moyamoya disease are major public health concerns. Accurate classification of these conditions is essential for timely intervention, yet current computer-aided methods often exhibit limited representational capacity, feature redundancy, and insufficient interpretability, restricting clinical applicability. We propose PASAformer, a Swin-Transformer-based framework for cerebrovascular disease classification on DSA. PASAformer incorporates a Pathology-Aware Sparse Attention (PASA) module that emphasizes lesion-related regions while suppressing background redundancy. Inserted into the Swin backbone, PASA replaces dense window self-attention, improving computational efficiency while preserving the hierarchical architecture. We further employ the MiAMix data augmenter to increase sample diversity, and incorporate a CombinedAdapter encoder that injects anatomical priors from the frozen Medical Segment Anything Model (MED-SAM) into early-stage representations, strengthening discriminative power under limited supervision. To support research in this underexplored area, we curate CDSA-NEO, a proprietary DSA dataset comprising more than 1,700 static images across four major cerebrovascular disease categories, constituting the first large-scale benchmark of its kind. Furthermore, an external cohort of angiographic runs with sequential, unselected frames is used to assess robustness in realistic temporal workflows. Extensive experiments on CDSA-NEO and public vascular datasets demonstrate that PASAformer achieves competitive precision and balanced accuracy compared to representative state-of-the-art models, while providing more focused visual explanations. These results suggest that PASAformer can support automated cerebrovascular disease classification on angiography, and that CDSA-NEO provides a benchmark for future method development and evaluation.

Autonomous vehicles must be comprehensively evaluated before deployed in cities and highways. However, most existing evaluation approaches for autonomous vehicles are static and lack adaptability, so they are usually inefficient in generating challenging scenarios for tested vehicles. In this paper, we propose an adaptive evaluation framework to efficiently evaluate autonomous vehicles in adversarial environments generated by deep reinforcement learning. Considering the multimodal nature of dangerous scenarios, we use ensemble models to represent different local optimums for diversity. We then utilize a nonparametric Bayesian method to cluster the adversarial policies. The proposed method is validated in a typical lane-change scenario that involves frequent interactions between the ego vehicle and the surrounding vehicles. Results show that the adversarial scenarios generated by our method significantly degrade the performance of the tested vehicles. We also illustrate different patterns of generated adversarial environments, which can be used to infer the weaknesses of the tested vehicles.

Safety is a critical concern when deploying reinforcement learning agents for realistic tasks. Recently, safe reinforcement learning algorithms have been developed to optimize the agent’s performance while avoiding violations of safety constraints. However, few studies have addressed the nonstationary disturbances in the environments, which may cause catastrophic outcomes. In this paper, we propose the context-aware safe reinforcement learning (CASRL) method, a metal-earning framework to realize safe adaptation in non-stationary environments. We use a probabilistic latent variable model to achieve fast inference of the posterior environment transition distribution given the context data. Safety constraints are then evaluated with uncertainty-aware trajectory sampling. Prior safety constraints are formulated with domain knowledge to improve safety during exploration. The algorithm is evaluated in realistic safety-critical environments with non-stationary disturbances. Results show that the proposed algorithm significantly outperforms existing baselines in terms of safety and robustness.

Long-tail and rare event problems become crucial when autonomous driving algorithms are applied in the real world. For the purpose of evaluating systems in challenging settings, we propose a generative framework to create safety-critical scenarios for evaluating specific task algorithms. We first represent the traffic scenarios with a series of autoregressive building blocks and generate diverse scenarios by sampling from the joint distribution of these blocks. We then train the generative model as an agent (or a generator) to search the risky scenario parameters for a given driving algorithm. We treat the driving algorithm as an environment that returns high reward to the agent when a risky scenario is generated. The whole process is optimized by the policy gradient reinforcement learning method. Through the experiments conducted on several scenarios in the simulation, we demonstrate that the proposed framework generates safety-critical scenarios more efficiently than grid search or human design methods. Another advantage of this method is its adaptiveness to the routes and parameters.

Multi-agent navigation in dynamic environments is of great industrial value when deploying a large scale fleet of robot to real-world applications. This paper proposes a decentralized partially observable multi-agent path planning with evolutionary reinforcement learning (MAPPER) method to learn an effective local planning policy in mixed dynamic environments. Reinforcement learning-based methods usually suffer performance degradation on long-horizon tasks with goal-conditioned sparse rewards, so we decompose the long-range navigation task into many easier sub-tasks under the guidance of a global planner, which increases agents' performance in large environments. Moreover, most existing multi-agent planning approaches assume either perfect information of the surrounding environment or homogeneity of nearby dynamic agents, which may not hold in practice. Our approach models dynamic obstacles' behavior with an image-based representation and trains a policy in mixed dynamic environments without homogeneity assumption. To ensure multi-agent training stability and performance, we propose an evolutionary training approach that can be easily scaled to large and complex environments. Experiments show that MAPPER is able to achieve higher success rates and more stable performance when exposed to a large number of non-cooperative dynamic obstacles compared with traditional reaction-based planner LRA* and the state-of-the-art learning-based method.

Continuously learning to solve unseen tasks with limited experience has been extensively pursued in meta-learning and continual learning, but with restricted assumptions such as accessible task distributions, independently and identically distributed tasks, and clear task delineations. However, real-world physical tasks frequently violate these assumptions, resulting in performance degradation. This paper proposes a continual online model-based reinforcement learning approach that does not require pre-training to solve task-agnostic problems with unknown task boundaries. We maintain a mixture of experts to handle nonstationarity, and represent each different type of dynamics with a Gaussian Process to efficiently leverage collected data and expressively model uncertainty. We propose a transition prior to account for the temporal dependencies in streaming data and update the mixture online via sequential variational inference. Our approach reliably handles the task distribution shift by generating new models for never-before-seen dynamics and reusing old models for previously seen dynamics. In experiments, our approach outperforms alternative methods in non-stationary tasks, including classic control with changing dynamics and decision making in different driving scenarios.

Interactions between vehicles and pedestrians have always been a major problem in traffic safety. Experienced human drivers are able to analyze the environment and choose driving strategies that will help them avoid crashes. What is not yet clear, however, is how automated vehicles will interact with pedestrians. This paper proposes a new method for evaluating the safety and feasibility of the driving strategy of automated vehicles when encountering unsignalized crossings. MobilEye® sensors installed on buses in Ann Arbor, Michigan, collected data on 2,973 valid crossing events. A stochastic interaction model was then created using a multivariate Gaussian mixture model. This model allowed us to simulate the movements of pedestrians reacting to an oncoming vehicle when approaching unsignalized crossings, and to evaluate the passing strategies of automated vehicles. A simulation was then conducted to demonstrate the evaluation procedure.