Tianyu Wo

Rui Wang, Chongwei Liu, Xudong Mou et al.

The accumulation of time-series data and the absence of labels make time-series Anomaly Detection (AD) a self-supervised deep learning task. Single-normality-assumption-based methods, which reveal only a certain aspect of the whole normality, are incapable of tasks involved with a large number of anomalies. Specifically, Contrastive Learning (CL) methods distance negative pairs, many of which consist of both normal samples, thus reducing the AD performance. Existing multi-normality-assumption-based methods are usually two-staged, firstly pre-training through certain tasks whose target may differ from AD, limiting their performance. To overcome the shortcomings, a deep Contrastive One-Class Anomaly detection method of time series (COCA) is proposed by authors, following the normality assumptions of CL and one-class classification. It treats the original and reconstructed representations as the positive pair of negative-sample-free CL, namely "sequence contrast". Next, invariance terms and variance terms compose a contrastive one-class loss function in which the loss of the assumptions is optimized by invariance terms simultaneously and the "hypersphere collapse" is prevented by variance terms. In addition, extensive experiments on two real-world time-series datasets show the superior performance of the proposed method achieves state-of-the-art.

Jinhao Chen, Zhen Yang, Jianxin Shi et al.

Multimodal large language models (MLLMs) have demonstrated remarkable capabilities in vision-language answering tasks. Despite their strengths, these models often encounter challenges in achieving complex reasoning tasks such as mathematical problem-solving. Previous works have focused on fine-tuning on specialized mathematical datasets. However, these datasets are typically distilled directly from teacher models, which capture only static reasoning patterns and leaving substantial gaps compared to student models. This reliance on fixed teacher-derived datasets not only restricts the model's ability to adapt to novel or more intricate questions that extend beyond the confines of the training data, but also lacks the iterative depth needed for robust generalization. To overcome these limitations, we propose \textbf{\method}, a \textbf{Math}ematical \textbf{S}elf-\textbf{E}volving framework for MLLMs. In contrast to traditional one-shot fine-tuning paradigms, \method iteratively refines the model through cycles of inference, reflection, and reward-based feedback. Specifically, we leverage iterative fine-tuning by incorporating correct reasoning paths derived from previous-stage inference and integrating reflections from a specialized Outcome Reward Model (ORM). To verify the effectiveness of \method, we evaluate it on a suite of challenging benchmarks, demonstrating significant performance gains over backbone models. Notably, our experimental results on MathVL-test surpass the leading open-source multimodal mathematical reasoning model QVQ. Our code and models are available at \texttt{https://zheny2751\allowbreak-dotcom.github.io/\allowbreak MathSE.github.io/}.

Xudong Mou, Rui Wang, Tiejun Wang et al.

Time series anomaly detection (TSAD) is a vital yet challenging task, particularly in scenarios where labeled anomalies are scarce and temporal dependencies are complex. Recent anomaly assumption (AA) approaches alleviate the lack of anomalies by injecting synthetic samples and training discriminative models. Despite promising results, these methods often suffer from two fundamental limitations: patchy generation, where scattered anomaly knowledge leads to overly simplistic or incoherent anomaly injection, and Anomaly Shift, where synthetic anomalies either resemble normal data too closely or diverge unrealistically from real anomalies, thereby distorting classification boundaries. In this paper, we propose CAPMix, a controllable anomaly augmentation framework that addresses both issues. First, we design a CutAddPaste mechanism to inject diverse and complex anomalies in a targeted manner, avoiding patchy generation. Second, we introduce a label revision strategy to adaptively refine anomaly labels, reducing the risk of anomaly shift. Finally, we employ dual-space mixup within a temporal convolutional network to enforce smoother and more robust decision boundaries. Extensive experiments on five benchmark datasets, including AIOps, UCR, SWaT, WADI, and ESA, demonstrate that CAPMix achieves significant improvements over state-of-the-art baselines, with enhanced robustness against contaminated training data. The code is available at https://github.com/alsike22/CAPMix.

Xiaohan Zhang, Xudong Mou, Rui Wang et al.

Redundant manipulators, with their higher Degrees of Freedom (DOFs), offer enhanced kinematic performance and versatility, making them suitable for applications like manufacturing, surgical robotics, and human-robot collaboration. However, motion planning for these manipulators is challenging due to increased DOFs and complex, dynamic environments. While traditional motion planning algorithms struggle with high-dimensional spaces, deep learning-based methods often face instability and inefficiency in complex tasks. This paper introduces RobotDiffuse, a diffusion model-based approach for motion planning in redundant manipulators. By integrating physical constraints with a point cloud encoder and replacing the U-Net structure with an encoder-only transformer, RobotDiffuse improves the model's ability to capture temporal dependencies and generate smoother, more coherent motion plans. We validate the approach using a complex simulator, and release a new dataset with 35M robot poses and 0.14M obstacle avoidance scenarios. Experimental results demonstrate the effectiveness of RobotDiffuse and the promise of diffusion models for motion planning tasks. The code can be accessed at https://github.com/ACRoboT-buaa/RobotDiffuse.

Zhuoren Ye, Tianyu Wo, Dinghao Xue et al.

Emerging LLM services increasingly host many sparse MoE models, yet most models receive sparse requests and remain cold. This creates a GPU memory problem: model weights are stable and model-determined, while KV-cache is transient and demand-determined. Because cold models rarely reach peak KV-cache demand at the same time, reserving worst-case KV capacity per model wastes memory; a shared KV-cache pool can instead provision aggregate active demand. However, KV-cache sharing is not sufficient when weights and KV-cache remain in a monolithic GPU memory pool. Static weights compete with dynamic KV-cache, and KV-head-limited attention under cold, low-concurrency traffic exposes only a fraction of replicated KV capacity, leading to low GPU memory utilization and weak long-context support. We present CrossPool, a serving engine for cold MoE models that separates FFN weights and KV-cache into two GPU memory pools: a weights pool that consolidates FFN weights across cold models, and a KV-cache pool that dynamically serves active requests while keeping attention local to KV-cache. CrossPool combines a KV-cache planner and virtualizer, a layer-wise pipeline scheduler that hides hidden-state transfers, and persistent kernels with control lowering to reduce CPU-GPU control overhead. With efficient GPU memory pooling, CrossPool underpins bursty long-context requests and outperforms the state-of-the-art kvcached-based multi-LLM serving system, reducing P99 TBT by up to $10.4\times$.

Jianxin Shi, Jinhao Chen, Yuandong Wang et al.

In recent years, the field of autonomous driving has attracted increasingly significant public interest. Accurately forecasting the future behavior of various traffic participants is essential for the decision-making of Autonomous Vehicles (AVs). In this paper, we focus on both scenario-based and perception-based motion forecasting for AVs. We propose a formal problem formulation for motion forecasting and summarize the main challenges confronting this area of research. We also detail representative datasets and evaluation metrics pertinent to this field. Furthermore, this study classifies recent research into two main categories: supervised learning and self-supervised learning, reflecting the evolving paradigms in both scenario-based and perception-based motion forecasting. In the context of supervised learning, we thoroughly examine and analyze each key element of the methodology. For self-supervised learning, we summarize commonly adopted techniques. The paper concludes and discusses potential research directions, aiming to propel progress in this vital area of AV technology.

Jinghao Wang, Xiao Zhou, Xiaoyang Sun et al.

Large Language Model based Multi-Agent Systems (LLM-MAS) have emerged as a powerful paradigm for tackling complex tasks by breaking them into collaborative workflows of specialized LLM-powered agents. However, deploying such multi-agent workloads at scale poses significant system challenges. Each user query spawns an iterative pipeline of LLM calls, greatly amplifying resource consumption compared to single-turn queries. In resource-constrained cloud settings, these workflows face non-deterministic and input-dependent costs at decode stage, heavy-tailed multi-model requirements with memory fragmentation and over-provisioning, and cross-cluster scheduling trade-offs. We present Maestro, a workload-aware scheduling system designed for LLM-MAS serving under strict GPU budgets. Maestro explicitly leverages agent semantics and roles: it predicts the output length and memory usage of each stage and uses this prediction to drive a hierarchical scheduler. At the node level, Maestro enables dynamic multi-model co-location via hierarchical weight caching and elastic memory provisioning. At the cluster level, it performs latency-aware routing to avoid cold-start delays and memory overloads. At the global level, it enforces workflow-aware prioritization to minimize head-of-line blocking for interactive tasks. Across prototype experiments and trace-driven simulations, Maestro reduces KV-reservation HBM by 67.2% and improves high-contention SLO attainment over EDF by 23.6 percentage points.

Yuandong Wang, Hongzhi Yin, Tong Chen et al.

In recent years, ride-hailing services have been increasingly prevalent as they provide huge convenience for passengers. As a fundamental problem, the timely prediction of passenger demands in different regions is vital for effective traffic flow control and route planning. As both spatial and temporal patterns are indispensable passenger demand prediction, relevant research has evolved from pure time series to graph-structured data for modeling historical passenger demand data, where a snapshot graph is constructed for each time slot by connecting region nodes via different relational edges (e.g., origin-destination relationship, geographical distance, etc.). Consequently, the spatiotemporal passenger demand records naturally carry dynamic patterns in the constructed graphs, where the edges also encode important information about the directions and volume (i.e., weights) of passenger demands between two connected regions. However, existing graph-based solutions fail to simultaneously consider those three crucial aspects of dynamic, directed, and weighted (DDW) graphs, leading to limited expressiveness when learning graph representations for passenger demand prediction. Therefore, we propose a novel spatiotemporal graph attention network, namely Gallat (Graph prediction with all attention) as a solution. In Gallat, by comprehensively incorporating those three intrinsic properties of DDW graphs, we build three attention layers to fully capture the spatiotemporal dependencies among different regions across all historical time slots. Moreover, the model employs a subtask to conduct pretraining so that it can obtain accurate results more quickly. We evaluate the proposed model on real-world datasets, and our experimental results demonstrate that Gallat outperforms the state-of-the-art approaches.