Robots perform diverse tasks in real-world scenarios. In safety-critical applications, robot control must prioritize satisfying safety constraints in addition to achieving high performance. Offline safe reinforcement learning avoids risky online exploration by training from a given dataset. However, most existing methods overlook two issues in offline data. First, non-zero cost signals are typically sparse, which leads to inaccurate cost value estimates and makes it difficult to impose effective safety constraints on the policy. Second, an imbalanced dataset biases policy learning toward unsafe behaviors. To address these challenges, we propose an actor-critic method ARMOR (multi-scAle Reweighting with Multi-task Offline cRitic). The multi-task critic treats reward, long-term cost, and short-term cost as multiple tasks, learns shared representations to capture common state information, and leverages dense reward signals to stabilize learning under sparse cost signals. To mitigate dataset imbalance, ARMOR performs counterfactual reasoning with the short-term cost to upweight critical safe transitions near the risk boundary and assigns higher weights to low-cost trajectories. It then performs multi-scale reweighting by combining transition-level and trajectory-level weights to debias data distribution and emphasize safe demonstrations. The actor is parameterized by a conditional diffusion policy and trained via weighted behavior cloning. ARMOR additionally incorporates a reward-guided objective and a long-term cost constraint to improve the reward-cost trade-off. Extensive experiments on continuous-control robot tasks show that ARMOR achieves competitive performance under safety constraints, with clear advantages in several challenging environments. Furthermore, ARMOR exhibits zero-shot adaptation capability, making it suitable for practical deployment.

Knowledge graphs (KGs) play a critical role in structured knowledge representation and downstream applications such as semantic search and intelligent decision-making. However, applying large language models (LLMs) to KG construction still faces challenges of generating hallucinated facts that undermine reliability and requiring costly fine-tuning to adapt to specialized domains. To address these issues, we propose TFreeKGGen, a training-free model for end-to-end KG generation. The model leverages hierarchical prompt design to fully exploit the semantic understanding of LLMs, and introduces a two-stage verification–correction mechanism that grounds extracted triples in source evidence and rectifies hallucinated outputs. Experiments on datasets from different domains demonstrate consistent improvement gained by TFreeKGGen. Moreover, we construct the first coalbed methane safety knowledge graph, demonstrating the model’s effectiveness in knowledge modeling. These results establish TFreeKGGen as an efficient and trustworthy pathway for scalable KG construction without domain-specific training, contributing to the advancement of knowledge engineering and intelligent decision support.

Time series anomaly detection is an important task in many applications, and deep learning based time series anomaly detection has made great progress. However, due to complex device interactions, time series exhibit diverse abnormal signal shapes, subtle anomalies, and imbalanced abnormal instances, which make anomaly detection in time series still a challenge. Fusion and analysis of multivariate time series can help uncover their intrinsic spatio-temporal characteristics, and contribute to the discovery of complex and subtle anomalies. In this paper, we propose a novel approach named Multi-scale Convolution Fusion and Memory-augmented Adversarial AutoEncoder (MCFMAAE) for multivariate time series anomaly detection. It is an encoder-decoder-based framework with four main components. Multi-scale convolution fusion module fuses multi-sensor signals and captures various scales of temporal information. Self-attention-based encoder adopts the multi-head attention mechanism for sequence modeling to capture global context information. Memory module is introduced to explore the internal structure of normal samples, capturing it into the latent space, and thus remembering the typical pattern. Finally, the decoder is used to reconstruct the signals, and then a process is coming to calculate the anomaly score. Moreover, an additional discriminator is added to the model, which enhances the representation ability of autoencoder and avoids overfitting. Experiments on public datasets demonstrate that MCFMAAE improves the performance compared to other state-of-the-art methods, which provides an effective solution for multivariate time series anomaly detection.

Long-term multivariate time series forecasting is an important task in engineering applications. It helps grasp the future development trend of data in real-time, which is of great significance for a wide variety of fields. Due to the non-linear and unstable characteristics of multivariate time series, the existing methods encounter difficulties in analyzing complex high-dimensional data and capturing latent relationships between multivariates in time series, thus affecting the performance of long-term prediction. In this paper, we propose a novel time series forecasting model based on multilayer perceptron that combines spatio-temporal decomposition and doubly residual stacking, namely Spatio-Temporal Decomposition Neural Network (STDNet). We decompose the originally complex and unstable time series into two parts, temporal term and spatial term. We design temporal module based on auto-correlation mechanism to discover temporal dependencies at the sub-series level, and spatial module based on convolutional neural network and self-attention mechanism to integrate multivariate information from two dimensions, global and local, respectively. Then we integrate the results obtained from the different modules to get the final forecast. Extensive experiments on four real-world datasets show that STDNet significantly outperforms other state-of-the-art methods, which provides an effective solution for long-term time series forecasting.

Recently, self-supervised learning has shown great potential in Graph Neural Networks (GNNs) through contrastive learning, which aims to learn discriminative features for each node without label information. The key to graph contrastive learning is data augmentation. The anchor node regards its augmented samples as positive samples, and the rest of the samples are regarded as negative samples, some of which may be positive samples. We call these mislabeled samples as "false negative" samples, which will seriously affect the final learning effect. Since such semantically similar samples are ubiquitous in the graph, the problem of false negative samples is very significant. To address this issue, the paper proposes a novel model, False negative sample Detection for Graph Contrastive Learning (FD4GCL), which uses attribute and structure-aware to detect false negative samples. Experimental results on seven datasets show that FD4GCL outperforms the state-of-the-art baselines and even exceeds several supervised methods.