The rapid advancement of Large Vision Language Models (LVLMs) has demonstrated excellent abilities in various visual tasks. Building upon these developments, the thinking with images paradigm has emerged, enabling models to dynamically edit and re-encode visual information at each reasoning step, mirroring human visual processing. However, this paradigm introduces significant challenges as diverse errors may occur during reasoning processes. This necessitates Process Reward Models (PRMs) for distinguishing positive and negative reasoning steps, yet existing benchmarks for PRMs are predominantly text-centric and lack comprehensive assessment under this paradigm. To address these gaps, this work introduces the first comprehensive benchmark specifically designed for evaluating PRMs under the thinking with images paradigm. Our main contributions are: (1) Through extensive analysis of reasoning trajectories and guided search experiments with PRMs, we define 7 fine-grained error types and demonstrate both the necessity for specialized PRMs and the potential for improvement. (2) We construct a comprehensive benchmark comprising 1,206 manually annotated thinking with images reasoning trajectories spanning 4 categories and 16 subcategories for fine-grained evaluation of PRMs. (3) Our experimental analysis reveals that current LVLMs fall short as effective PRMs, exhibiting limited capabilities in visual reasoning process evaluation with significant performance disparities across error types, positive evaluation bias, and sensitivity to reasoning step positions. These findings demonstrate the effectiveness of our benchmark and establish crucial foundations for advancing PRMs in LVLMs.

Chi-Min Chan, Yujin Zhou, Pengcheng Wen, Boqin Yin, Jiaming Ji, Juntao Dai, Wei Xue, Sirui Han, Yike Guo. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Han Zhu, Juntao Dai, Jiaming Ji, Haoran Li, Chengkun Cai, Pengcheng Wen, Chi-Min Chan, Boyuan Chen, Yaodong Yang, Sirui Han, Yike Guo. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

High-speed railway operations are frequently disrupted by both external and internal factors. This study addresses train rescheduling and passenger reassignment in high-speed railway systems with seat reservation mechanisms by considering both operator and passenger perspectives and incorporating passenger compensation strategies to enhance service quality. The impacts of disruptions are quantified in terms of train delay costs and passenger compensation costs. An integrated mixed-integer programming (MIP) model is developed to minimize the weighted total operating cost, subject to constraints on train operations, station and track capacities, train capacities, and passenger boarding, transfer, and compensation rules. To solve the proposed model, a BD-RH computational framework is introduced by combining Benders decomposition (BD) and the rolling horizon (RH) algorithm, which decomposes the integrated problem into a train rescheduling master problem and a passenger reassignment subproblem for iterative optimization. Numerical experiments on both an artificial network and a subnetwork of China’s high-speed railway system validate the effectiveness and efficiency of the proposed approach. The findings provide actionable guidance for formulating passenger compensation strategies and passenger-oriented rescheduling strategies.

Cross-domain facial expression recognition is confronted by the problem of the large distribution discrepancy and samples inconsistencies between the source domain and target domain. To solve this problem, we propose a cross-domain sample relationship learning (CSRL) method that explores useful intrinsic sample relationships of two domains to narrow the domain discrepancy. Specifically, during the training stage, we first design inter-domain sample transformers to explore the sample similarity relationships between the source and target domains, and then deploy intra-domain sample transformers to capture the internal similar structure of the samples in each domain. Thus dual sample relationships can be learned to align the cross-domain similar samples and preserve the domain-specific information, which can facilitate both the inter-domain invariant features and intra-domain invariant features learning. Subsequently, we design a joint alignment strategy by simultaneously deploying the feature distribution alignment and cross-domain sample relationship learning. Thus, both local similar samples and global domain distribution of two domains can be well aligned to enhance the generalization ability of the model. Experimental results on several benchmark databases show the superiority of CSRL over some state-of-the-art methods.

Speech emotion recognition has lots of applications such as human‐computer interaction and health management. The current methods are challenged with the problems of fuzzy decision boundary and imbalance between difficult and easy samples in the training data. This paper first proposes an additive angle penalty focus loss function (APFL), which strictly refines the fuzzy decision boundary by introducing angle penalty factors to improve the compactness within the class and enlarge the distance between classes. It also assigns the larger loss to difficult samples to make the model pay more attention to them, as they are easily misclassified. Simultaneously, due to the lack of training samples, the framework of multimodal and multitask learning with APFL is further proposed, which extracts spectrogram features by deep neural network, text features by the pretrained language model, and audio features by the pretrained sound model. It uses the gender recognition as an auxiliary task. The experimental results verify the effectiveness of the proposed loss function and framework.

Railway operations are vulnerable to unexpected disruptions that should be handled in an efficient and passenger-friendly way. This paper focuses on a real-time train timetable rescheduling problem in a railway system with a seat-reserved mechanism during disruptions. An integer linear programming model with a compensation mechanism is established to provide a better overall solution for dynamic and multistage rescheduling problems. The main part of the model considers the strategies of retiming, adding train stops and reordering to reduce the impact of passengers, while the compensation mechanism part mainly deals with the uncertainty of disruption durations. The related constraints mainly include section track capacity, station track capacity,passenger transfer rules and some robust measures under the uncertainty of disruption durations. A multistage decision optimization framework is established, and a rolling horizon algorithm is embedded in it to approximate the desired optimal solution. A series of numerical experiments based on a real case of China’s high-speed railway subnetwork are carried out to verify the effectiveness and efficiency of the proposed optimization approach.

The utility to decode hand movement parameters is significant to the control of artificial limb in the BCI fields. Most previous studies have adopted amplitude features of the low-frequency EEG signals to decode hand movement parameters. In this study, we have investigated the instantaneous phase of the low-frequency EEG signals attained by Hilbert transform for such a task for the first time, and compared its decoding accuracy with that of the amplitude features. An experiment was carried out that 5 subjects executed a center-out reaching task in two sessions. Then the Multiple Linear Regression (MLR) model is used to decode hand movement parameters based on the amplitude feature and the phase feature, respectively. The performance of the proposed approach is evaluated by calculating the correlation coefficients between the recorded parameters and the reconstructed parameters. The experiments results show that compared to the decoder with the amplitude feature, the correlation coefficients obtained by the decoder with the phase feature have increased 27.8% (X-position), 24.1% (Y-position), 27.9% (X-velocity), 20.9% (Y-velocity).