HiFloat4 Format for Language Model Pre-training on Ascend NPUs

Large foundation models have become central to modern machine learning, with performance scaling predictably with model size and data. However, training and deploying such models incur substantial computational and memory costs, motivating the development of low-precision training techniques. Recent work has demonstrated that 4-bit floating-point (FP4) formats--such as MXFP4 and NVFP4--can be successfully applied to linear GEMM operations in large language models (LLMs), achieving up to 4x improvements in compute throughput and memory efficiency compared to higher-precision baselines. In this work, we investigate the recently proposed HiFloat4 FP4 format for Huawei Ascend NPUs and systematically compare it with MXFP4 in large-scale training settings. All experiments are conducted on Ascend NPU clusters, with linear and expert GEMM operations performed entirely in FP4 precision. We evaluate both dense architectures (e.g., Pangu and LLaMA-style models) and mixture-of-experts (MoE) models, where both standard linear layers and expert-specific GEMMs operate in FP4. Furthermore, we explore stabilization techniques tailored to FP4 training that significantly reduce numerical degradation, maintaining relative error within 1% of full-precision baselines while preserving the efficiency benefits of 4-bit computation. Our results provide a comprehensive empirical study of FP4 training on NPUs and highlight the practical trade-offs between FP4 formats in large-scale dense and MoE models.

SPROUT: A Scalable Diffusion Foundation Model for Agricultural Vision

Vision Foundation Models (VFM) pre-trained on large-scale unlabeled data have achieved remarkable success on general computer vision tasks, yet typically suffer from significant domain gaps when applied to agriculture. In this context, we introduce $SPROUT$ ($S$calable $P$lant $R$epresentation model via $O$pen-field $U$nsupervised $T$raining), a multi-crop, multi-task agricultural foundation model trained via diffusion denoising. SPROUT leverages a VAE-free Pixel-space Diffusion Transformer to learn rich, structure-aware representations through denoising and enabling efficient end-to-end training. We pre-train SPROUT on a curated dataset of 2.6 million high-quality agricultural images spanning diverse crops, growth stages, and environments. Extensive experiments demonstrate that SPROUT consistently outperforms state-of-the-art web-pretrained and agricultural foundation models across a wide range of downstream tasks, while requiring substantially lower pre-training cost. The code and model are available at https://github.com/UTokyo-FieldPhenomics-Lab/SPROUT.

Terahertz Beam Squint Mitigation via Six-Dimensional Movable Antennas

Analog beamforming holds great potential for future terahertz (THz) communications due to its ability to generate high-gain directional beams with low-cost phase shifters. However, conventional analog beamforming may suffer substantial performance degradation in wideband systems due to the beam squint effect. Instead of relying on high-cost true-time delayers, we propose an efficient six-dimensional movable antenna (6DMA) architecture to mitigate the beam-squint effect. In particular, we study a wideband wide-beam coverage problem in this paper, aiming to maximize the minimum beamforming gain over a given range of azimuth/elevation angles and frequencies by jointly optimizing the analog beamforming vector, the MA positions within a two-dimensional (2D) region, and the three-dimensional (3D) rotation angles of the antenna array. However, this problem is non-convex and intractable to solve optimally due to the coupling of the spatial and frequency domains and that of the antenna weights, positions and rotation. To tackle this problem, we first derive an optimal solution to it in a special case with azimuth or elevation angle coverage only. It is shown that rotating a uniform linear array (ULA) is sufficient to achieve global optimality and eliminate beam-squint effects. While for other general cases, an alternating optimization (AO) algorithm is proposed to obtain a high-quality suboptimal solution, where the antennas' beamforming weights, positions, and rotation angles are alternately optimized by combining successive convex approximation (SCA), sequential update with Gibbs sampling (GS), and hybrid coarse- and fine-grained search. Simulation results demonstrate that our proposed scheme can significantly outperform conventional antenna arrays without antenna movement or rotation, thus offering a cost-effective solution for wideband transmission over THz bands.

SpecSteer: Synergizing Local Context and Global Reasoning for Efficient Personalized Generation

Realizing personalized intelligence faces a core dilemma: sending user history to centralized large language models raises privacy concerns, while on-device small language models lack the reasoning capacity required for high-quality generation. Our pilot study shows that purely local enhancements remain insufficient to reliably bridge this gap. We therefore propose SpecSteer, an asymmetric collaborative inference framework that synergizes private on-device context with cloud-scale reasoning. SpecSteer casts collaboration as Bayesian knowledge fusion and repurposes speculative decoding as a distributed alignment protocol, yielding a Draft--Verify--Recover pipeline: the on-device model drafts personalized sequences; the cloud validates via a ratio-based mechanism that decouples reasoning verification from private context, filtering logical flaws without accessing raw user context; upon rejection, a steering recovery injects local intent during correction. Experiments demonstrate that SpecSteer successfully closes the reasoning gap and achieves superior personalized generation performance, while delivering a 2.36x speedup over standard baselines.

Rethinking Mutual Coupling in Movable Antenna MIMO Systems

Movable antenna (MA) systems have emerged as a promising technology for future wireless communication systems. The movement of antennas gives rise to mutual coupling (MC) effects, which have been previously ignored and can be exploited to enhance the capacity of multiple-input multiple-output (MIMO) systems. To this end, we first model an MA-enabled point-to-point MIMO communication system with MC effects using a circuit-theoretic framework. The capacity maximization problem is then formulated as a non-concave optimization problem and solved via a block coordinate ascent (BCA)-based algorithm. The subproblem of optimizing MA positions is challenging due to the presence of the analytically intractable MC matrices. To overcome this difficulty, we develop a trust region method (TRM)-based algorithm to optimize MA positions, wherein Sylvester equations are employed to compute the derivatives of the inverse square roots of the MC matrices. Simulation results show significant capacity gains from leveraging MC effects, primarily due to customizable MC matrices and superdirectivity.

GuiDINO: Rethinking Vision Foundation Model in Medical Image Segmentation

Foundation vision models are increasingly adopted in medical image analysis. Due to domain shift, these pretrained models misalign with medical image segmentation needs without being fully fine-tuned or lightly adapted. We introduce GuiDINO, a framework that repositions native foundation model to acting as a visual guidance generator for downstream segmentation. GuiDINO extracts visual feature representation from DINOv3 and converts them into a spatial guide mask via a lightweight TokenBook mechanism, which aggregates token-prototype similarities. This guide mask gates feature activations in multiple segmentation backbones, thereby injecting foundation-model priors while preserving the inductive biases and efficiency of medical dedicated architectures. Training relies on a guide supervision objective loss that aligns the guide mask to ground-truth regions, optionally augmented by a boundary-focused hinge loss to sharpen fine structures. GuiDINO also supports parameter-efficient adaptation through LoRA on the DINOv3 guide backbone. Across diverse medical datasets and nnUNet-style inference, GuiDINO consistently improves segmentation quality and boundary robustness, suggesting a practical alternative to fine-tuning and offering a new perspective on how foundation models can best serve medical vision. Code is available at https://github.com/Hi-FishU/GuiDINO

WirelessAgent++: Automated Agentic Workflow Design and Benchmarking for Wireless Networks

The integration of large language models (LLMs) into wireless networks has sparked growing interest in building autonomous AI agents for wireless tasks. However, existing approaches rely heavily on manually crafted prompts and static agentic workflows, a process that is labor-intensive, unscalable, and often suboptimal. In this paper, we propose WirelessAgent++, a framework that automates the design of agentic workflows for various wireless tasks. By treating each workflow as an executable code composed of modular operators, WirelessAgent++ casts agent design as a program search problem and solves it with a domain-adapted Monte Carlo Tree Search (MCTS) algorithm. Moreover, we establish WirelessBench, a standardized multi-dimensional benchmark suite comprising Wireless Communication Homework (WCHW), Network Slicing (WCNS), and Mobile Service Assurance (WCMSA), covering knowledge reasoning, code-augmented tool use, and multi-step decision-making. Experiments demonstrate that \wap{} autonomously discovers superior workflows, achieving test scores of $78.37\%$ (WCHW), $90.95\%$ (WCNS), and $97.07\%$ (WCMSA), with a total search cost below $\$ 5$ per task. Notably, our approach outperforms state-of-the-art prompting baselines by up to $31\%$ and general-purpose workflow optimizers by $11.1\%$, validating its effectiveness in generating robust, self-evolving wireless agents. The code is available at https://github.com/jwentong/WirelessAgent-R2.

FuXi-Linear: Unleashing the Power of Linear Attention in Long-term Time-aware Sequential Recommendation

Modern recommendation systems primarily rely on attention mechanisms with quadratic complexity, which limits their ability to handle long user sequences and slows down inference. While linear attention is a promising alternative, existing research faces three critical challenges: (1) temporal signals are often overlooked or integrated via naive coupling that causes mutual interference between temporal and semantic signals while neglecting behavioral periodicity; (2) insufficient positional information provided by existing linear frameworks; and (3) a primary focus on short sequences and shallow architectures. To address these issues, we propose FuXi-Linear, a linear-complexity model designed for efficient long-sequence recommendation. Our approach introduces two key components: (1) a Temporal Retention Channel that independently computes periodic attention weights using temporal data, preventing crosstalk between temporal and semantic signals; (2) a Linear Positional Channel that integrates positional information through learnable kernels within linear complexity. Moreover, we demonstrate that FuXi-Linear exhibits a robust power-law scaling property at a thousand-length scale, a characteristic largely unexplored in prior linear recommendation studies. Extensive experiments on sequences of several thousand tokens demonstrate that FuXi-Linear outperforms state-of-the-art models in recommendation quality, while achieving up to 10$\times$ speedup in the prefill stage and up to 21$\times$ speedup in the decode stage compared to competitive baselines. Our code has been released in a public repository https://github.com/USTC-StarTeam/fuxi-linear.

Generative Data Transformation: From Mixed to Unified Data

Recommendation model performance is intrinsically tied to the quality, volume, and relevance of their training data. To address common challenges like data sparsity and cold start, recent researchs have leveraged data from multiple auxiliary domains to enrich information within the target domain. However, inherent domain gaps can degrade the quality of mixed-domain data, leading to negative transfer and diminished model performance. Existing prevailing \emph{model-centric} paradigm -- which relies on complex, customized architectures -- struggles to capture the subtle, non-structural sequence dependencies across domains, leading to poor generalization and high demands on computational resources. To address these shortcomings, we propose \textsc{Taesar}, a \emph{data-centric} framework for \textbf{t}arget-\textbf{a}lign\textbf{e}d \textbf{s}equenti\textbf{a}l \textbf{r}egeneration, which employs a contrastive decoding mechanism to adaptively encode cross-domain context into target-domain sequences. It employs contrastive decoding to encode cross-domain context into target sequences, enabling standard models to learn intricate dependencies without complex fusion architectures. Experiments show \textsc{Taesar} outperforms model-centric solutions and generalizes to various sequential models. By generating enriched datasets, \textsc{Taesar} effectively combines the strengths of data- and model-centric paradigms. The code accompanying this paper is available at~ \textcolor{blue}{https://github.com/USTC-StarTeam/Taesar}.

Unlearning Noise in PINNs: A Selective Pruning Framework for PDE Inverse Problems

Physics-informed neural networks (PINNs) provide a promising framework for solving inverse problems governed by partial differential equations (PDEs) by integrating observational data and physical constraints in a unified optimization objective. However, the ill-posed nature of PDE inverse problems makes them highly sensitive to noise. Even a small fraction of corrupted observations can distort internal neural representations, severely impairing accuracy and destabilizing training. Motivated by recent advances in machine unlearning and structured network pruning, we propose P-PINN, a selective pruning framework designed to unlearn the influence of corrupted data in a pretrained PINN. Specifically, starting from a PINN trained on the full dataset, P-PINN evaluates a joint residual--data fidelity indicator, a weighted combination of data misfit and PDE residuals, to partition the training set into reliable and corrupted subsets. Next, we introduce a bias-based neuron importance measure that quantifies directional activation discrepancies between the two subsets, identifying neurons whose representations are predominantly driven by corrupted samples. Building on this, an iterative pruning strategy then removes noise-sensitive neurons layer by layer. The resulting pruned network is fine-tuned on the reliable data subject to the original PDE constraints, acting as a lightweight post-processing stage rather than a complete retraining. Numerical experiments on extensive PDE inverse-problem benchmarks demonstrate that P-PINN substantially improves robustness, accuracy, and training stability under noisy conditions, achieving up to a 96.6\% reduction in relative error compared with baseline PINNs. These results indicate that activation-level post hoc pruning is a promising mechanism for enhancing the reliability of physics-informed learning in noise-contaminated settings.