arXiv subjects

Yuhao Feng

Publications and source records attributed to Yuhao Feng.

8 recordsLinked to original sources

arXiv2026

SkyReels-V4: Multi-modal Video-Audio Generation, Inpainting and Editing model

SkyReels V4 is a unified multi modal video foundation model for joint video audio generation, inpainting, and editing. The model adopts a dual stream Multimodal Diffusion Transformer (MMDiT) architecture, where one branch synthesizes video and the other generates temporally aligned audio, while sharing a powerful text encoder based on the Multimodal Large Language Models (MLLM). SkyReels V4 accepts rich multi modal instructions, including text, images, video clips, masks, and audio references. By combining the MLLMs multi modal instruction following capability with in context learning in the video branch MMDiT, the model can inject fine grained visual guidance under complex conditioning, while the audio branch MMDiT simultaneously leverages audio references to guide sound generation. On the video side, we adopt a channel concatenation formulation that unifies a wide range of inpainting style tasks, such as image to video, video extension, and video editing under a single interface, and naturally extends to vision referenced inpainting and editing via multi modal prompts. SkyReels V4 supports up to 1080p resolution, 32 FPS, and 15 second duration, enabling high fidelity, multi shot, cinema level video generation with synchronized audio. To make such high resolution, long-duration generation computationally feasible, we introduce an efficiency strategy: Joint generation of low resolution full sequences and high-resolution keyframes, followed by dedicated super-resolution and frame interpolation models. To our knowledge, SkyReels V4 is the first video foundation model that simultaneously supports multi-modal input, joint video audio generation, and a unified treatment of generation, inpainting, and editing, while maintaining strong efficiency and quality at cinematic resolutions and durations.

cs.CV↗

arXiv2025

High-Efficiency Octave Bandwidth Rectifier for Electromagnetic Energy Harvesting

This letter presents the design and implementation of a compact high-efficiency octave microwave rectifier. A key highlight is the novel segmented impedance matching method, a unique approach that expands the rectifier bandwidth. The diode reactance is initially regulated by a series short-ended microstrip line. Impedance-compensated structures, characterized by varying admittance properties across an extensive frequency range, partition the operating frequency band into two segments based on the input impedance, thereby minimizing impedance variation. Ultimately, the input impedance is matched by a novel triple-band matching network. An octave rectifier was fabricated and measured. Results demonstrate that the rectifier achieves over 50% efficiency over 1.3-2.55 GHz fractional bandwidth 64.9% at 0-dBm RF input power. Even with a decrease in input power to -10 dBm, the rectifier maintains over 30% efficiency.

physics.app-ph↗

arXiv2025

Design of a High-Power and High-Efficiency GaN-HEMT VCO Based on an Inverse Class-F Amplifier

This letter proposes a high-power and high-efficiency GaN-HEMT voltage-controlled oscillator. The VCO consists of a coupled-line coupler, an inverse class-F amplifier, and a novel frequency-tunable stepped-impedance resonator. Using a harmonic control circuit and a parasitic parameter compensation circuit, the power amplifier operates in the inverse class-F state to achieve high efficiency. The feedback circuit uses a coupled-line coupler instead of the traditional coupling capacitor to control feedback power precisely. The measurement results show that the VCO with an oscillation frequency of 2.41-2.45 GHz achieves a maximum conversion efficiency of 74.5% at 2.44 GHz and an output power of 40.2 dBm. It is a candidate for the microwave source in a wireless power transmission system.

physics.app-ph↗

arXiv2025

A Novel Class-F 2.45/5.8 GHz Dual-Band Rectifier for Wireless Power Transmission

This letter proposes a high-efficiency dual-band class-F rectifier for wireless power transmission (WPT). The rectifier comprises a dual-band harmonic termination network, a dual-band matching network, a single Schottky diode, and a dc pass filter. A theoretical analysis of the harmonic termination network is performed to improve the rectifying efficiency. The network exhibits good class-F operation by controlling the second and third harmonics in dual bands. A rectifier operating at 2.45 and 5.8 GHz was designed, fabricated, and measured for validation. The measurements show maximum RF-dc conversion efficiencies of 74.9% and 61.9% with 200 and 500{\Omega} loads at 2.45 and 5.8 GHz, respectively. The proposed rectifier achieves dual-band harmonic control with high efficiency.

physics.app-ph↗

arXiv2025

Wideband Low-Scattering Dual-Polarized Phased Array with Stepped Ground

This paper proposes a wideband dual-polarized phased array with ultra-wideband scattering cross section (SCS) reduction. The antenna elements are loaded on a bilateral stepped ground. This ground is carefully designed in terms of height difference, step number, and length to achieve phase cancellation near the normal direction. Wideband dipoles with vertical electric coupling are designed. The radiation frequency band covers the X-band (40%) under VSWR < 2.8. Array patterns are synthesized with the two subarrays, covering the scanning range from -45 to +45 degrees. The monostatic SCSs of the proposed 17 x 8 array prototype have been reduced within 3.6 - 30 GHz, with an averaged reduction of over 19.4/18.9 dB and an averaged in-band reduction of over 15.4/16.6 dB, under the normal x/y polarized incident waves respectively.

physics.optics↗

arXiv2025

Global Patterns of Extreme Temperature Teleconnections Using Climate Network Analysis

Extreme weather events, rare yet profoundly impactful, are often accompanied by severe conditions. Increasing global temperatures are poised to exacerbate these events, resulting in greater human casualties, economic losses, and ecological destruction. Complex global climate interactions, known as teleconnections, can lead to widespread repercussions triggered by localized extreme weather. Understanding these teleconnection patterns is crucial for weather forecasting, enhancing safety, and advancing climate science. Here, we employ climate network analysis to uncover teleconnection patterns associated with extreme temperature fluctuations, including both extreme warming and cooling events occurring on a daily basis. Our study results demonstrate that the distances of significant teleconnections initially conform to a power-law decay, signifying a decline in connectivity with distance. However, this power-law decay tendency breaks beyond a certain threshold distance, suggesting the existence of long-distance connections. Additionally, we uncover a greater prevalence of long-distance connectivity among extreme cooling events compared to extreme warming events. The global pattern of teleconnections is, in part, driven by the mechanism of Rossby waves, which serve as a rapid conduit for inducing correlated fluctuations in both pressure and temperature. These results enhance our understanding of the multiscale nature of climate teleconnections and hold significant implications for improving weather forecasting and assessing climate risks in a warming world.

physics.ao-ph↗

arXiv2024

Baichuan4-Finance Technical Report

Large language models (LLMs) have demonstrated strong capabilities in language understanding, generation, and reasoning, yet their potential in finance remains underexplored due to the complexity and specialization of financial knowledge. In this work, we report the development of the Baichuan4-Finance series, including a comprehensive suite of foundational Baichuan4-Finance-Base and an aligned language model Baichuan4-Finance, which are built upon Baichuan4-Turbo base model and tailored for finance domain. Firstly, we have dedicated significant effort to building a detailed pipeline for improving data quality. Moreover, in the continual pre-training phase, we propose a novel domain self-constraint training strategy, which enables Baichuan4-Finance-Base to acquire financial knowledge without losing general capabilities. After Supervised Fine-tuning and Reinforcement Learning from Human Feedback and AI Feedback, the chat model Baichuan4-Finance is able to tackle various financial certification questions and real-world scenario applications. We evaluate Baichuan4-Finance on many widely used general datasets and two holistic financial benchmarks. The evaluation results show that Baichuan4-Finance-Base surpasses almost all competitive baselines on financial tasks by significant margins without sacrificing performance on general LLM benchmarks. At the same time, Baichuan4-Finance demonstrates even more impressive performance on financial application scenarios, showcasing its potential to foster community innovation in the financial LLM field.

cs.CL↗

arXiv2024

Dynamic data sampler for cross-language transfer learning in large language models

Large Language Models (LLMs) have gained significant attention in the field of natural language processing (NLP) due to their wide range of applications. However, training LLMs for languages other than English poses significant challenges, due to the difficulty in acquiring large-scale corpus and the requisite computing resources. In this paper, we propose ChatFlow, a cross-language transfer-based LLM, to address these challenges and train large Chinese language models in a cost-effective manner. We employ a mix of Chinese, English, and parallel corpus to continuously train the LLaMA2 model, aiming to align cross-language representations and facilitate the knowledge transfer specifically to the Chinese language model. In addition, we use a dynamic data sampler to progressively transition the model from unsupervised pre-training to supervised fine-tuning. Experimental results demonstrate that our approach accelerates model convergence and achieves superior performance. We evaluate ChatFlow on popular Chinese and English benchmarks, the results indicate that it outperforms other Chinese models post-trained on LLaMA-2-7B.

cs.CL↗