Emotional Support Conversation (ESC) plays a critical role in mental health assistance by providing accessible psychological support in real-world applications. Large Language Models (LLMs) have shown strong empathetic abilities in ESC tasks. Yet, existing methods overlook the issue of cognitive distortions in help-seekers’ expressions. As a result, current models can only provide basic emotional comfort, rather than helping help-seekers address their psychological distress at a deeper cognitive level. To address this challenge, we construct the CogBiasESC dataset, the first dataset that expands existing ESC datasets by adding labels for cognitive distortions, includes their type, intensity, and safe risk level. Furthermore, we propose the Cognitive Policy-driven Large Language Model framework (CoPoLLM) to enhance LLMs’ ability to diagnose and intervene cognitive distortions in help-seekers. We also analyze the safety advantages of CoPoLLM from a theoretical perspective. Experimental results show that CoPoLLM significantly outperforms 15 state-of-the-art baselines in terms of distortion diagnosis accuracy, intervention strategy effectiveness, and safety risk control. Our source code is available at: https://github.com/Chips98/CoPoLLM-for-ACL-2026.

Recently, Heterogeneous Graph Contrastive Learning (HGCL) has received significant attention due to its impressive capability to represent heterogeneous graphs without detailed annotations. However, the inherent fragility of heterogeneous graph structures makes HGCL vulnerable to perturbation attacks. Most existing defense works for heterogeneous graphs primarily focus on supervised scenarios, which protect all nodes equally via structural pruning. This defensive mechanism can result in insufficient structure information for HGCL, thus degrading performance in self-supervised scenarios without labels. In this paper, we argue that some nodes are more susceptible to attacks, and the influence of the perturbation attack will accumulate across layers during representation aggregation. To tackle these problems, we propose a novel Heterogeneous Graph Contrastive Learning with Hierarchical Vulnerability Protection (HVP-HGCL), which identifies the most vulnerable nodes to perturbation attack and protects them across different aggregation layers to improve the robustness of HGCL. Specifically, we first design the Vulnerability Detection (VD) based on the HGCL framework to determine which nodes are more sensitive to attack in self-supervised scenarios. Subsequently, we propose a simple but efficient Hierarchical Protection (HP) to safeguard those vulnerable nodes from attack noise during different layers. Combining the above two modules, HVP-HGCL can not only improve the robustness of HGCL but also ensure sufficient structural information for effective contrastive learning. Extensive experiments demonstrate that HVP-HGCL improves robustness against adversarial attacks and achieves competitive performance on downstream tasks.

Multi-task prompt learning has emerged as a promising technique for fine-tuning pre-trained Vision-Language Models (VLMs) to various downstream tasks. However, existing methods ignore challenges caused by spurious correlations and dynamic task relationships, which may reduce the model performance. To tackle these challenges, we propose JSCPT, a novel approach for Joint Scheduling of Causal Prompts and Tasks to enhance multi-task prompt learning. Specifically, we first design a Multi-Task Vison-Language Prompt (MTVLP) model, which learns task-shared and task-specific vison-language prompts and selects useful prompt features via causal intervention, alleviating spurious correlations. Then, we propose the task-prompt scheduler that models inter-task affinities and assesses the causal effect of prompt features to optimize the multi-task prompt learning process. Finally, we formulate the scheduler and the multi-task prompt learning process as a bi-level optimization problem to optimize prompts and tasks adaptively. In the lower optimization, MTVLP is updated with the scheduled gradient, while in the upper optimization, the scheduler is updated with the implicit gradient. Extensive experiments show the superiority of our proposed JSCPT approach over several baselines in terms of multi-task prompt learning for pre-trained VLMs.

Current parameter-efficient fine-tuning (PEFT) methods have shown superior performance in continual learning. However, most existing PEFT-based methods focus on mitigating catastrophic forgetting by limiting modifications to the old task model caused by new tasks. This hinders backward knowledge transfer, as when new tasks have a strong positive correlation with old tasks, appropriately training on new tasks can transfer beneficial knowledge to old tasks. Critically, achieving backward knowledge transfer faces two fundamental challenges: (1) some parameters may be ineffective on task performance, which constrains the task solution space and model capacity; (2) since old task data are inaccessible, modeling task correlation via shared data is infeasible. To address these challenges, we propose CaLoRA, a novel \textbf{c}ausal-\textbf{a}ware \textbf{lo}w-\textbf{r}ank \textbf{a}daptation framework that is the first PEFT-based continual learning work with backward knowledge transfer. Specifically, we first propose \textbf{p}ar\textbf{a}meter-level \textbf{c}ounterfactual \textbf{a}ttribution (PaCA) that estimates the causal effect of LoRA parameters via counterfactual reasoning, identifying effective parameters from a causal view. Second, we propose \textbf{c}ross-t\textbf{a}sk \textbf{g}radient \textbf{a}daptation (CaGA) to quantify task correlation by gradient projection and evaluate task affinity based on gradient similarity. By incorporating causal effect, task correlation, and affinity, CaGA adaptively adjusts task gradients, facilitating backward knowledge transfer without relying on data replay. Extensive experiments across multiple benchmarks and continual learning settings show that CaLoRA outperforms state-of-the-art methods. In particular, CaLoRA better mitigates catastrophic forgetting by enabling positive backward knowledge transfer.

Recent studies reveal that Large Language Models (LLMs) exhibit strong sequential reasoning capabilities, allowing them to replace specialized time-series models and serve as foundation models for complex time-series analysis. To activate the capabilities of LLMs for time-series tasks, numerous studies have attempted to bridge the gap between time series and linguistics by aligning textual representations with time-series patterns. However, it is a non-trivial endeavor to losslessly capture the infinite time-domain variability using natural language, leading to suboptimal alignment performance. Beyond representation, contextual differences, where semantics in time series are conveyed by consecutive points, unlike in text by individual tokens, are often overlooked by existing methods. To address these, we propose S$^2$TS-LLM, a simple yet effective framework to repurpose LLMs for universal time series analysis through the following two main paradigms: (i) a spectral symbolization paradigm transforms time series into frequency-domain representations characterized by a fixed number of components and prominent amplitudes, which enables a limited set of symbols to effectively abstract key frequency features; (ii) a contextual segmentation paradigm partitions the sequence into blocks based on temporal patterns and reassigns positional encodings accordingly, thereby mitigating the structural mismatch between time series and natural language. Together, these paradigms bootstrap the LLMs' perception of temporal patterns and structures, effectively bridging time series and linguistics. Extensive experiments show that S$^2$TS-LLM can serve as a powerful time series analyzer, outperforming state-of-the-art methods across time series tasks.

Stance detection on social media aims to identify if an individual is in support of or against a specific target. Most existing stance detection approaches primarily rely on modeling the contextual semantic information in sentences and neglect to explore the pragmatics dependency information of words, thus degrading performance. Although several single-task learning methods have been proposed to capture richer semantic representation information, they still suffer from semantic sparsity problems caused by short texts on social media. This article proposes a novel multigraph sparse interaction network (MG-SIN) by using multitask learning (MTL) to identify the stances and classify the sentiment polarities of tweets simultaneously. Our basic idea is to explore the pragmatics dependency relationship between tasks at the word level by constructing two types of heterogeneous graphs, including task-specific and task-related graphs (tr-graphs), to boost the learning of task-specific representations. A graph-aware module is proposed to adaptively facilitate information sharing between tasks via a novel sparse interaction mechanism among heterogeneous graphs. Through experiments on two real-world datasets, compared with the state-of-the-art baselines, the extensive results exhibit that MG-SIN achieves competitive improvements of up to 2.1% and 2.42% for the stance detection task, and 5.26% and 3.93% for the sentiment analysis task, respectively.

Graph contrastive learning (GCL) has gained increasing interest as a solution for graph representation learning. In GCL, graph augmentation is essential to generate contrastive samples used for contrastive learning. Recently, most existing methods employ learnable graph view generators to augment graphs based on the node probability distribution adaptively. However, these methods cannot ensure that semantic-related nodes are preserved during graph augmentation, leading to performance degradation. To tackle this issue, we propose a novel approach called Semantic-aware Graph Contrastive Learning (SGCL), which can generate high-quality contrastive samples by only augmenting semantic-unrelated nodes so as to facilitate the performance of GCL on downstream tasks. Specifically, we first design a Lipschitz constant generator to compute the Lipschitz constants that measure the semantic relevance of each node. Then, we propose the Lipschitz graph augmentation to augment graphs while only dropping these semantic-unrelated nodes with small Lipschitz constants. Furthermore, we propose semanticaware contrastive learning to obtain more refined representations by contrasting the graph-level representation of anchor graphs and high-quality generated samples. Experimental results on unsupervised learning and transfer learning demonstrate the effectiveness of SGCL compared to state-of-the-art methods.

Multi-task learning (MTL) has emerged as a promising approach for sharing inductive bias across multiple tasks to enable more efficient learning in text classification.However, training all tasks simultaneously often yields degraded performance of each task than learning them independently, since different tasks might conflict with each other.Existing MTL methods for alleviating this issue is to leverage heuristics or gradient-based algorithm to achieve an arbitrary Pareto optimal trade-off among different tasks.In this paper, we present a novel gradient trade-off approach to mitigate the task conflict problem, dubbed GetMTL, which can achieve a specific tradeoff among different tasks nearby the main objective of multi-task text classification (MTC), so as to improve the performance of each task simultaneously.The results of extensive experiments on two benchmark datasets back up our theoretical analysis and validate the superiority of our proposed GetMTL.

Molecular classification benefits a lot from the re-cent success of graph contrastive learning (GCL) which pulls positive samples close and pushes the negative samples apart. GCL methods generate negative and positive samples via graph augmentation. Due to the structural corruption caused by graph augmentation, not all generated negative samples retain discrim-inative semantics. However, existing GCL methods ignore the difference between negative samples and hold an assumption that the importance of all negative samples is the same, leading to degraded performance of molecular classification. To address this issue, in this paper, we propose a novel molecular graph contrastive learning model (MocGCL) by selecting more useful negative samples to improve the performance of molecular classification. Specifically, we first employ different encoders to generate positive samples to improve the diversity of positive samples. Then, we design negative generation to generate negative samples and define semantic integrity to measure the usefulness of generated negative samples. Moreover, we propose the novel negative selection to dynamically select the negative samples of more usefulness to improve the molecular representation. In addition, we improve the contrastive loss to adaptively adjust the distance between selected negative samples, which can pre-serve the distinctive properties of selected negative samples in sample space. Extensive experiments on six typical bioinformatics datasets demonstrate the effectiveness of our MocGCL compared to most state-of-the-art methods.

Point cloud completion is an interesting and challenging task in 3D vision, which aims to recover complete shapes from sparse and incomplete point clouds. Existing completion networks often require a vast number of parameters and substantial computational costs to achieve a high performance level, which may limit their practical application. In this work, we propose a novel Adaptive efficient Recurrent Forward Network (ARFNet), which is composed of three parts: Recurrent Feature Extraction (RFE), Forward Dense Completion (FDC) and Raw Shape Protection (RSP). In an RFE, multiple short global features are extracted from incomplete point clouds, while a dense quantity of completed results are generated in a coarse-to-fine pipeline in the FDC. Finally, we propose the Adamerge module to preserve the details from the original models by merging the generated results with the original incomplete point clouds in the RSP. In addition, we introduce the Sampling Chamfer Distance to better capture the shapes of the models and the balanced expansion constraint to restrict the expansion distances from coarse to fine. According to the experiments on ShapeNet and KITTI, our network can achieve state-of-the-art completion performances on dense point clouds with fewer parameters, smaller model sizes, lower memory costs and a faster convergence.

Stance detection aims to identify people's standpoints expressed in the text towards a target, which can provide powerful information for various downstream tasks.Recent studies have proposed multi-task learning models that introduce sentiment information to boost stance detection.However, they neglect to explore capturing the fine-grained task-specific interaction between stance detection and sentiment tasks, thus degrading performance.To address this issue, this paper proposes a novel multi-task interaction network (MTIN) for improving the performance of stance detection and sentiment analysis tasks simultaneously.Specifically, we construct heterogeneous taskrelated graphs to automatically identify and adapt the roles that a word plays with respect to a specific task.Also, a multi-task interaction module is designed to capture the wordlevel interaction between tasks, so as to obtain richer task representations.Extensive experiments on two real-world datasets show that our proposed approach outperforms state-ofthe-art methods in both stance detection and sentiment analysis tasks.

Point cloud completion is an interesting and challenging task in 3D vision, aiming to recover complete shapes from sparse and incomplete point clouds. Existing learning-based methods often require vast computation cost to achieve excellent performance, which limits their practical applications. In this paper, we propose a novel Recurrent Forward Network (RFNet), which is composed of three modules: Recurrent Feature Extraction (RFE), Forward Dense Completion (FDC) and Raw Shape Protection (RSP). The RFE extracts multiple global features from the incomplete point clouds for different recurrent levels, and the FDC generates point clouds in a coarse-to-fine pipeline. The RSP introduces details from the original incomplete models to refine the completion results. Besides, we propose a Sampling Chamfer Distance to better capture the shapes of models and a new Balanced Expansion Constraint to restrict the expansion distances from coarse to fine. According to the experiments on ShapeNet and KITTI, our network can achieve the state-of-the-art with lower memory cost and faster convergence.

In this paper, we present PocoNet: Point cloud Online COmpression NETwork to address the task of SLAM-oriented compression. The aim of this task is to select a compact subset of points with high priority to maintain localization accuracy. The key insight is that points with high priority have similar geometric features in SLAM scenarios. Hence, we tackle this task as point cloud segmentation to capture complex geometric information. We calculate observation counts by matching between maps and point clouds and divide them into different priority levels. Trained by labels annotated with such observation counts, the proposed network could evaluate the point-wise priority. Experiments are conducted by integrating our compression module into an existing SLAM system to evaluate compression ratios and localization performances. Experimental results on two different datasets verify the feasibility and generalization of our approach.

Multi-agent path finding in formation has many potential real-world applications like mobile warehouse robotics. However, previous multi-agent path finding (MAPF) methods hardly take formation into consideration. Further-more, they are usually centralized planners and require the whole state of the environment. Other decentralized partially observable approaches to MAPF are reinforcement learning (RL) methods. However, these RL methods encounter difficulties when learning path finding and formation problems at the same time. In this paper, we propose a novel decentralized partially observable RL algorithm that uses a hierarchical structure to decompose the multi-objective task into unrelated ones. It also calculates a theoretical weight that makes each tasks reward has equal influence on the final RL value function. Additionally, we introduce a communication method that helps agents cooperate with each other. Experiments in simulation show that our method outperforms other end-to-end RL methods and our method can naturally scale to large world sizes where centralized planner struggles. We also deploy and validate our method in a real-world scenario.

This paper presents F-Siamese Tracker, a novel approach for single object tracking prominently characterized by more robustly integrating 2D and 3D information to reduce redundant search space. A main challenge in 3D single object tracking is how to reduce search space for generating appropriate 3D candidates. Instead of solely relying on 3D proposals, firstly, our method leverages the Siamese network applied on RGB images to produce 2D region proposals which are then extruded into 3D viewing frustums. Besides, we perform an on-line accuracy validation on the 3D frustum to generate refined point cloud searching space, which can be embedded directly into the existing 3D tracking backbone. For efficiency, our approach gains better performance with fewer candidates by reducing search space. In addition, benefited from introducing the online accuracy validation, for occasional cases with strong occlusions or very sparse points, our approach can still achieve high precision, even when the 2D Siamese tracker loses the target. This approach allows us to set a new state-of-the-art in 3D single object tracking by a significant margin on a sparse outdoor dataset (KITTI tracking). Moreover, experiments on 2D single object tracking show that our framework boosts 2D tracking performance as well.