Multivariate time series forecasting is essential in domains such as finance, transportation, climate, and energy. However, existing patch-based methods typically adopt fixed-length segmentation, overlooking the heterogeneity of local temporal dynamics and the decoding heterogeneity of forecasting. Such designs lose details in information-dense regions, introduce redundancy in stable segments, and fail to capture the distinct complexities of short-term and long-term horizons. We propose TimeMosaic, a forecasting framework that aims to address temporal heterogeneity. TimeMosaic employs adaptive patch embedding to dynamically adjust granularity according to local information density, balancing motif reuse with structural clarity while preserving temporal continuity. In addition, it introduces segment-wise decoding that treats each prediction horizon as a related subtask and adapts to horizon-specific difficulty and information requirements, rather than applying a single uniform decoder. Extensive evaluations on benchmark datasets demonstrate that TimeMosaic delivers consistent improvements over existing methods, and our model trained on the large-scale corpus with 321 billion observations achieves performance competitive with state-of-the-art TSFMs.

Reliable and interpretable electrocardiogram (ECG) analysis has long been a grand challenge in medical artificial intelligence, requiring complex reasoning, ethical awareness, and multimodal understanding. Despite recent progress in large language models (LLMs), three major limitations persist. First, existing ECG datasets primarily focus on static classification or single-turn QA, lacking support for interactive and context-dependent reasoning. Second, current evaluation frameworks overlook ethical and patient-centered perspectives, failing to capture responsibility and safety in diagnostic reasoning. Third, real-world diagnostic diversity and data imbalance remain underrepresented, limiting model robustness and generalization. To address these issues, we present ECG-Expert-QA, a unified multimodal benchmark for diagnostic reasoning and conversational evaluation of medical LLMs. It integrates real-world ECG records with systematically synthesized cases, covering 12 diagnostic tasks and 47,211 expert-validated QA pairs. The benchmark uniquely supports multi-turn doctor-patient dialogues, ethical risk assessment, and cross-modal clinical reasoning, enabling comprehensive evaluation across reasoning depth and contextual understanding. We further conduct extensive benchmarking of mainstream medical LLMs (GPT-4o, DeepSeek-V3, Qwen2.5) and fine-tuning experiments on the lightweight MiniMind2, demonstrating that ECG-Expert-QA effectively differentiates model capabilities in semantic comprehension, clinical reasoning, and ethical expression. Built upon a knowledge-guided and roledriven generation framework, ECG-Expert-QA ensures linguistic precision and clinical fidelity. The dataset has been fully opensourced on GitHub (available at: https://github.com/Zaozzz/ECG-Expert-QA), providing a reproducible foundation for developing trustworthy, explainable, and patient-centered medical LLMs.

Multivariate Time Series Forecasting plays a key role in many applications. Recent works have explored using Large Language Models for MTSF to take advantage of their reasoning abilities. However, many methods treat LLMs as end-to-end forecasters, which often leads to a loss of numerical precision and forces LLMs to handle patterns beyond their intended design. Alternatively, methods that attempt to align textual and time series modalities within latent space frequently encounter alignment difficulty. In this paper, we propose to treat LLMs not as standalone forecasters, but as semantic guidance modules within a dual-stream framework. We propose DualSG, a dual-stream framework that provides explicit semantic guidance, where LLMs act as Semantic Guides to refine rather than replace traditional predictions. As part of DualSG, we introduce Time Series Caption, an explicit prompt format that summarizes trend patterns in natural language and provides interpretable context for LLMs, rather than relying on implicit alignment between text and time series in the latent space. We also design a caption-guided fusion module that explicitly models inter-variable relationships while reducing noise and computation. Experiments on real-world datasets from diverse domains show that DualSG consistently outperforms 15 state-of-the-art baselines, demonstrating the value of explicitly combining numerical forecasting with semantic guidance.

Remote sensing data from hyperspectral imaging (HSI) and LiDAR provide complementary perspectives for terrain and object analysis. However, existing methods for multimodal data fusion primarily focus on spatial-domain feature alignment, often overlooking the potential of frequency-domain information to enhance classification accuracy. To bridge this gap, we introduce the Frequency-Domain Hierarchical Perception Fusion Network (FHPF-Net), a novel framework for precise classification of remote sensing images. This network leverages both spatial and frequency-domain information, and provides a new perspective for heterogeneous data integration. To extract and utilize frequency-domain features, we propose the HighLow Spectral Separation and Mining (HLSSM) module, which isolates high-frequency details such as edges and textures from low-frequency structural patterns in HSI and LiDAR data. This separation facilitates targeted feature extraction while preserving crucial contextual information. Additionally, we introduce the Hierarchical Superimposed Multi-domain Information Fusion (HSMIF) module, which employs a multi-level fusion strategy to integrate spatial and frequency-domain features, ensuring consistency and complementarity between the two data sources. Finally, we introduce a Learnable Voting Pre-label Fusion (LVPF) strategy to effectively integrate multi-branch outputs, enhancing classification performance and model robustness. The proposed FHPF-Net effectively captures diverse responses across heterogeneous data types, enabling robust classification in complex environments. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art methods.

Video outpainting aims to complete missing areas at the edges of video frames adequately. Compared to image outpainting, it presents an additional challenge as the model should maintain the temporal consistency of the filled area. In this paper, we introduce a masked 3D diffusion model for video outpainting. We use the technique of mask modeling to train the 3D diffusion model. This allows us to use multiple guide frames to connect the results of multiple video clip inferences, thus ensuring temporal consistency and reducing jitter between adjacent frames. Meanwhile, we extract the global frames of the video as prompts and guide the model to obtain information other than the current video clip using cross-attention. We also introduce a hybrid coarse-to-fine inference pipeline to alleviate the artifact accumulation problem. The existing coarse-to-fine pipeline only uses the infilling strategy, which brings degradation because the time interval of the sparse frames is too large. Our pipeline benefits from bidirectional learning of the mask modeling and thus can employ a hybrid strategy of infilling and interpolation when generating sparse frames. Experiments show that our method achieves state-of-art results in video outpainting tasks.

Earlier-stage evaluations of a new AI architecture/system need affordable AI benchmarks. Only using a few AI component benchmarks like MLPerf alone in the other stages may lead to misleading conclusions. Moreover, the learning dynamics are not well understood, and the benchmarks' shelf-life is short. This paper proposes a balanced benchmarking methodology. We use real-world benchmarks to cover the factors space that impacts the learning dynamics to the most considerable extent. After performing an exhaustive survey on Internet service AI domains, we identify and implement nineteen representative AI tasks with state-of-the-art models. For repeatable performance ranking (RPR subset) and workload characterization (WC subset), we keep two subsets to a minimum for affordability. We contribute by far the most comprehensive AI training benchmark suite. The evaluations show: (1) AIBench Training (v1.1) outperforms MLPerf Training (v0.7) in terms of diversity and representativeness of model complexity, computational cost, convergent rate, computation, and memory access patterns, and hotspot functions; (2) Against the AIBench full benchmarks, its RPR subset shortens the benchmarking cost by 64%, while maintaining the primary workload characteristics; (3) The performance ranking shows the single-purpose AI accelerator like TPU with the optimized TensorFlow framework performs better than that of GPUs while losing the latter's general support for various AI models. The specification, source code, and performance numbers are available from the AIBench homepage https://www.benchcouncil.org/aibench-training/index.html.