Mingjun Zhang

Expert-as-a-Service: Towards Efficient, Scalable, and Robust Large-scale MoE Serving

Ziming Liu, Boyu Tian, Guoteng Wang, Zhen Jiang et autres

Mixture-of-Experts (MoE) models challenge serving infrastructures with dynamic, sparse expert utilization, causing instability on conventional systems designed for dense architectures. We propose EaaS, a novel serving system to enable efficient, scalable, and robust MoE deployment. Our system disaggregates MoE modules into independent, …

Higher-order HDL: Applied to MLP neural network hardware implementation

Samuel Garcia, Mingjun Zhang

In this article, we describe a methodology for the rapid implementation of a hardware architecture using a higher-order approach. This methodology uses a combination of TCL and VHDL for higher-order coding (i.e. code produced by code) and is supported by industry-standard HDL …

Domain-weighted transfer learning and discriminative embeddings for low-resource speaker verification

Han Wang, Mingrui He, Mingjun Zhang, Changzhi Luo et autres

Transfer learning has been shown to be effective in enhancing speaker verification performance in low-resource conditions. However, the inclusion of additional datasets may cause domain mismatch. Additionally, mismatched data volume and model complexity during fine-tuning can degrade speaker verification performance. In this …

cn (code pays fourni par la source)

Non-Target Conversion Based Speech Steganography for Secure Speech Communication System

Mingjun Zhang, Yan Feng, Yu Gao, Longting Xu

The widespread application of speech data increases the risk of speaker identity being compromised during speech communication. To mitigate this risk and protect voice privacy, we propose a system aimed at ensuring the security of speech communication by avoiding the exposure of …

cn (code pays fourni par la source)

ASM-GAN: Arbitrary Object Shape Styles Manipulation in Synthesized Images for GANs

Wei Li, Mingjun Zhang, Yinghui Wang, Xiaowu Zhang et autres

Utilizing deep learning-based algorithms to perform vision-based measurement (VBM) tasks is in the ascendant, especially for detecting defectives (i.e., objects in our study) in the industrial product inspection domain. The deep learning algorithms heavily rely on sufficiently labeled defective samples with different …

cn, bd (code pays fourni par la source)

Evaluation of genotype characteristics and drug resistance mutations in patients with chronic hepatitis B

Changlong He, Xiaoli Wu, Zhonglan You, Tao Zhou et autres

Hepatitis B is one of the public health priorities worldwide, especially in the Southwest China. Our study aimed to investigate the relationship between genotypes and drug resistance mutations among HBV patients in Southwest China, with the objective of providing guidance for clinical …

cn (code pays fourni par la source)

A neural circuit for lavender-essential-oil-induced antinociception

Yumeng Yang, Hao Huang, Mengyu Zhu, Hong-Rui Wei et autres

neurons abolishes LEO-induced antinociception. Our study thus depicts an LEO-responsive olfactory system circuit mechanism for alleviating inflammatory pain via aPir→IC neural connections, providing evidence to support development of aroma-based treatments for alleviating pain.

cn (code pays fourni par la source)

Device Feature based on Graph Fourier Transformation with Logarithmic Processing For Detection of Replay Speech Attacks

Mingrui He, Longting Xu, Han Wang, Mingjun Zhang et autres

Device Feature based on Graph Fourier Transformation with Logarithmic Processing For Detection of Replay Speech Attacks

Mingrui He, Longting Xu, Han Wang, Mingjun Zhang et autres

The most common spoofing attacks on automatic speaker verification systems are replay speech attacks. Detection of replay speech heavily relies on replay configuration information. Previous studies have shown that graph Fourier transform-derived features can effectively detect replay speech but ignore device and …

Evaluation of genotype characteristics and drug resistance mutations in patients with chronic hepatitis B

Changlong He, Xiaoli Wu, Zhonglan You, Liping Diao et autres

cn (code pays fourni par la source)

Research on library construction driven by multi-technology integration

Beibei Xu, V. A. Kuz'minykh, Shiwei Zhu, Junfeng Yu et autres

The deep mining of data and the application of technology integration are the main characteristics of a smart society. This paper is to explore the application of multi-technology integration in library smart construction in a smart society, and study the library construction …

cn, ua (code pays fourni par la source)

NTF4 plays a dual role in breast cancer in mammary tumorigenesis and metastatic progression

Ran Sun, Jin He, Xiang Qin, Yixiao Feng et autres

Breast cancer metastasis can happen even when the primary tumor is relatively small.But the mechanism for such early metastasis is poorly understood.Herein, we report that neurotrophin 4 (NTF4) plays a dual role in breast cancer proliferation and metastasis.Clinical data showed high levels …

cn (code pays fourni par la source)