Long-form science fiction generation demands rigorous maintenance of narrative coherence across evolving plots, character dynamics, and speculative world-building. We propose Octopus, an entropy-controlled neural framework with persistent memory-context binding that addresses these challenges through two key innovations: 1) dynamic entropy regulation balancing creativity and structural stability via narrative divergence thresholds, and 2) hierarchical memory architecture preserving character states, plot events, and scientific rules over 10K+ token spans. Evaluations across 12 sci-fi subgenres demonstrate Octopus's superiority over GPT-4 and ReAlign baselines, achieving 15.2% higher coherence scores (SciClarity) and 62% fewer contextual contradictions in extended narratives. Human evaluations confirm its effectiveness in maintaining speculative logic (4.7/5 vs. 3.1/5 baseline) while preserving creative diversity. The framework resolves the "hard sci-fi paradox" of enforcing scientific rigor without compromising narrative flexibility, establishing new capabilities for AI-assisted cross-media universe development.

Low Earth Orbit (LEO) satellite constellations have experienced rapid growth in recent years, driven by their potential to deliver global, high-bandwidth Internet services with low latency. Beyond connectivity, LEO constellations also offer promising opportunities to enable in-orbit processing of space-native data to support a wide range of emerging space applications. In this context, the concept of space computing has been proposed, a paradigm that seamlessly integrates networking and computing to provide computing-as-a-service anytime and anywhere in space. However, the inherent characteristics of satellite constellations, such as dynamic network topologies, constrained system resources, and the harsh space environment, pose significant challenges in achieving this vision. This paper outlines the system architecture and the key enabling technologies for space computing, including spaceborne computers, laser communications, spaceborne router, distributed operating systems, and onboard AI. We also present the implementation of an open space computing platform, the 3-Body computing constellation, along with the in-orbit experimental results that demonstrate the advantages of multi-satellite distributed computing. Furthermore, we outline future research directions essential for advancing toward a truly interconnected, autonomous, and intelligent space computing system.

In this work, we focus on a novel and practical task, i.e., Time-vAriant iMage inPainting (TAMP). The aim of TAMP is to restore a damaged target image by leveraging the complementary information from a reference image, where both images capture the same scene but with a significant time gap in between, i.e., time-variant images. Different from conventional reference-guided image inpainting, the reference image under TAMP setup presents significant content distinction to the target image and potentially also suffers from damages. Such an application frequently happens in our daily life to restore a damaged image by referring to another reference image, where there is no guarantee of the reference image's source and quality. In particular, our study finds that even SOTA reference-guided image inpainting methods fail to achieve plausible results due to the chaotic image complementation. To address such an ill-posed problem, we propose a novel Interactive Distribution Transition Estimation (InDiTE) module which interactively complements the time-variant images with appropriate semantics thus facilitate the restoration of damaged regions. To further boost the performance, we propose our TAMP solution, namely Interactive Distribution Transition Estimation-driven Diffusion (InDiTE-Diff), which integrates InDiTE with SOTA diffusion model and conducts latent cross-reference during sampling. Moreover, considering the lack of benchmarks for TAMP task, we newly assembled a dataset, i.e., TAMP-Street, based on existing image and mask datasets. We conduct experiments on the TAMP-Street datasets under two different time-variant image inpainting settings, which show our method consistently outperform SOTA reference-guided image inpainting methods for solving TAMP.

The high cost and low success rate of de novo drug development have spurred growing interest in drug repurposing. Computational methods leverage large-scale biomedical knowledge graphs (KGs) to discover new indications for existing drugs. However, structural heterogeneity, relation sparsity, and limited interpretability remain major challenges. We propose RiemannGNN, an end-to-end framework integrating hyperbolic geometry, graph neural networks (GNNs), and causal inference for interpretable drug repurposing. RiemannGNN embeds multirelational biomedical KGs into a hyperbolic manifold modeled by the Poincar'e ball, effectively capturing hierarchical biomedical structures. On top of this geometry, a causal reasoning module extracts disease-centered subgraphs and performs interventionbased inference via structural causal models, yielding mechanistic explanations for predicted drug-disease links. Evaluated on PrimeKG, RiemannGNN significantly outperforms state-of-the-art baselines. In a case study on amyotrophic lateral sclerosis (ALS), it identifies 20 novel candidate drugs with interpretable causal paths explaining their therapeutic relevance. Several predictions unsupported in the KG are validated by literature, while others suggest new hypotheses for future study. These results highlight the promise of combining hyperbolic representation learning with causal reasoning for robust and explainable biomedical AI.

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.

Humans naturally understand moments in a video by integrating visual and auditory cues. For example, localizing a scene in the video like “A scientist passionately speaks on wildlife conservation as dramatic orchestral music plays, with the audience nodding and applauding” requires simultaneous processing of visual, audio, and speech signals. However, existing models often struggle to effectively fuse and interpret audio information, limiting their capacity for comprehensive video temporal understanding. To address this, we present TriSense, a triple-modality large language model designed for holistic video temporal understanding through the integration of visual, audio, and speech modalities. Central to TriSense is a Query-Based Connector that adaptively reweights modality contributions based on the input query, enabling robust performance under modality dropout and allowing flexible combinations of available inputs. To support TriSense's multimodal capabilities, we introduce TriSense-2M, a high-quality dataset of over 2 million curated samples generated via an automated pipeline powered by fine-tuned LLMs. TriSense-2M includes long-form videos and diverse modality combinations, facilitating broad generalization. Extensive experiments across multiple benchmarks demonstrate the effectiveness of TriSense and its potential to advance multimodal video analysis.

In unmanned aerial systems, especially in complex environments, accurately detecting tiny objects is crucial. Resizing images is a common strategy to improve detection accuracy, particularly for small objects. However, simply enlarging images significantly increases computational costs and the number of negative samples, severely degrading detection performance and limiting its applicability. This paper proposes a Dynamic Pooling Network (DPNet) for tiny object detection to mitigate these issues. DPNet employs a flexible down-sampling strategy by introducing a factor (df) to relax the fixed down-sampling process of the feature map to an adjustable one. Furthermore, we design a lightweight predictor to predict df for each input image, which will be used to decrease the resolution of feature map in backbone. Thus, we achieve input-aware down-sampling. We design an Adaptive Normalization Module (ANM) to make a unified detector well compatible with different dfs. At the same time, we also design a guidance loss to supervise the predictor’s training. DPNet realizes the dynamic allocation of computing resources to trade off detection accuracy and efficiency through this. Experiments on the TinyCOCO and TinyPerson datasets show that our DPNet can save over 35% and 25% GFLOPs, respectively, while maintaining comparable detection performance.The code will be made publicly available.

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.

In high-throughput intelligent computing scenarios, multi-device parallelism strategies based on data parallelism or pipeline parallelism have been extensively utilized to accelerate large deep neural network model inference. Data parallelism offers nearly linear improvement in inference speed, but it is limited by the memory capacity of a single device which constrains the model size. On the other hand, pipeline parallelism can support larger models, but the total communication of the activations among devices is high, which limits the improvement of the inference speed. To address the demand for efficient model inference in high-throughput heterogeneous scenarios, we proposes a hybrid parallelism strategy that combines data parallelism and pipeline parallelism. The strategy involves grouping heterogeneous device clusters and then employing inter-group data parallelism along with intra-group pipeline parallelism. Moreover, we propose an algorithm to find an optimal hybrid parallel inference strategy with maximum throughput. The control variables of the strategy includes the number of groups, group-device assignments and model partition ratios. Our experimental evaluation demonstrates that compared to PipeEdge, a pipeline parallel inference framework for heterogeneous cluster, our strategy can achieve 1.7× −3.4× acceleration in an 8-device heterogeneous cluster without loss of accuracy.