Search

Your search keyword '"Lu,Shun"' showing total 3,105 results
Start Over Author "Lu,Shun"
3,105 results on '"Lu,Shun"'

1. PID: Physics-Informed Diffusion Model for Infrared Image Generation

Author

Mao, Fangyuan, Mei, Jilin, Lu, Shun, Liu, Fuyang, Chen, Liang, Zhao, Fangzhou, and Hu, Yu

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Infrared imaging technology has gained significant attention for its reliable sensing ability in low visibility conditions, prompting many studies to convert the abundant RGB images to infrared images. However, most existing image translation methods treat infrared images as a stylistic variation, neglecting the underlying physical laws, which limits their practical application. To address these issues, we propose a Physics-Informed Diffusion (PID) model for translating RGB images to infrared images that adhere to physical laws. Our method leverages the iterative optimization of the diffusion model and incorporates strong physical constraints based on prior knowledge of infrared laws during training. This approach enhances the similarity between translated infrared images and the real infrared domain without increasing extra training parameters. Experimental results demonstrate that PID significantly outperforms existing state-of-the-art methods. Our code is available at https://github.com/fangyuanmao/PID.

Published
2024

17. Construction of a prognostic model for gastric cancer based on immune infiltration and microenvironment, and exploration of MEF2C gene function

Author

Si-yu Wang, Yu-xin Wang, Lu-shun Guan, Ao Shen, Run-jie Huang, Shu-qiang Yuan, Yu-long Xiao, Li-shuai Wang, Dan Lei, Yin Zhao, Chuan Lin, Chang-ping Wang, and Zhi-ping Yuan

Subjects
Gastric cancer (GC), Prognostic model, Transcriptome sequencing, Immune microenvironment, Immunotherapy, Transcription factor (TF), Internal medicine, RC31-1245, Genetics, QH426-470
Abstract

Abstract Background Advanced gastric cancer (GC) exhibits a high recurrence rate and a dismal prognosis. Myocyte enhancer factor 2c (MEF2C) was found to contribute to the development of various types of cancer. Therefore, our aim is to develop a prognostic model that predicts the prognosis of GC patients and initially explore the role of MEF2C in immunotherapy for GC. Methods Transcriptome sequence data of GC was obtained from The Cancer Genome Atlas (TCGA), the Gene Expression Omnibus (GEO) and PRJEB25780 cohort for subsequent immune infiltration analysis, immune microenvironment analysis, consensus clustering analysis and feature selection for definition and classification of gene M and N. Principal component analysis (PCA) modeling was performed based on gene M and N for the calculation of immune checkpoint inhibitor (ICI) Score. Then, a Nomogram was constructed and evaluated for predicting the prognosis of GC patients, based on univariate and multivariate Cox regression. Functional enrichment analysis was performed to initially investigate the potential biological mechanisms. Through Genomics of Drug Sensitivity in Cancer (GDSC) dataset, the estimated IC50 values of several chemotherapeutic drugs were calculated. Tumor-related transcription factors (TFs) were retrieved from the Cistrome Cancer database and utilized our model to screen these TFs, and weighted correlation network analysis (WGCNA) was performed to identify transcription factors strongly associated with immunotherapy in GC. Finally, 10 patients with advanced GC were enrolled from Sun Yat-sen University Cancer Center, including paired tumor tissues, paracancerous tissues and peritoneal metastases, for preparing sequencing library, in order to perform external validation. Results Lower ICI Score was correlated with improved prognosis in both the training and validation cohorts. First, lower mutant-allele tumor heterogeneity (MATH) was associated with lower ICI Score, and those GC patients with lower MATH and lower ICI Score had the best prognosis. Second, regardless of the T or N staging, the low ICI Score group had significantly higher overall survival (OS) compared to the high ICI Score group. For its mechanisms, consistently, for Camptothecin, Doxorubicin, Mitomycin, Docetaxel, Cisplatin, Vinblastine, Sorafenib and Paclitaxel, all of the IC50 values were significantly lower in the low ICI Score group compared to the high ICI Score group. As a result, based on univariate and multivariate Cox regression, ICI Score was considered to be an independent prognostic factor for GC. And our Nomogram showed good agreement between predicted and actual probabilities. Based on CIBERSORT deconvolution analysis, there was difference of immune cell composition found between high and low ICI Score groups, probably affecting the efficacy of immunotherapy. Then, MEF2C, a tumor-related transcription factor, was screened out by WGCNA analysis. Higher MEF2C expression is significantly correlated with a worse OS. Moreover, its higher expression is also negatively correlated with tumor mutation burden (TMB) and microsatellite instability (MSI), but positively correlated with several immunosuppressive molecules, indicating MEF2C may exert its influence on tumor development by upregulating immunosuppressive molecules. Finally, based on transcriptome sequencing data on 10 paired tumor tissues from Sun Yat-sen University Cancer Center, MEF2C expression was significantly lower in paracancerous tissues compared to tumor tissues and peritoneal metastases, and it was also lower in tumor tissues compared to peritoneal metastases, indicating a potential positive association between MEF2C expression and tumor invasiveness. Conclusions Our prognostic model can effectively predict outcomes and facilitate stratification GC patients, offering valuable insights for clinical decision-making. The identified transcription factor MEF2C can serve as a biomarker for assessing the efficacy of immunotherapy for GC.

Published
2025
Full Text
View/download PDF

18. Repotrectinib in ROS1 Fusion-Positive Non-Small-Cell Lung Cancer.

Author

Drilon, Alexander, Camidge, D, Lin, Jessica, Kim, Sang-We, Solomon, Benjamin, Dziadziuszko, Rafal, Besse, Benjamin, Goto, Koichi, de Langen, Adrianus, Wolf, Jürgen, Lee, Ki, Popat, Sanjay, Springfeld, Christoph, Nagasaka, Misako, Felip, Enriqueta, Yang, Nong, Velcheti, Vamsidhar, Lu, Shun, Kao, Steven, Dooms, Christophe, Krebs, Matthew, Yao, Wenxiu, Beg, Muhammad, Hu, Xiufeng, Moro-Sibilot, Denis, Cheema, Parneet, Stopatschinskaja, Shanna, Mehta, Minal, Trone, Denise, Graber, Armin, Sims, Gregory, Yuan, Yong, and Cho, Byoung

Subjects
Humans, Carcinoma, Non-Small-Cell Lung, Lung Neoplasms, Protein-Tyrosine Kinases, Proto-Oncogene Proteins, Antineoplastic Agents, Treatment Outcome
Abstract

BACKGROUND: The early-generation ROS1 tyrosine kinase inhibitors (TKIs) that are approved for the treatment of ROS1 fusion-positive non-small-cell lung cancer (NSCLC) have antitumor activity, but resistance develops in tumors, and intracranial activity is suboptimal. Repotrectinib is a next-generation ROS1 TKI with preclinical activity against ROS1 fusion-positive cancers, including those with resistance mutations such as ROS1 G2032R. METHODS: In this registrational phase 1-2 trial, we assessed the efficacy and safety of repotrectinib in patients with advanced solid tumors, including ROS1 fusion-positive NSCLC. The primary efficacy end point in the phase 2 trial was confirmed objective response; efficacy analyses included patients from phase 1 and phase 2. Duration of response, progression-free survival, and safety were secondary end points in phase 2. RESULTS: On the basis of results from the phase 1 trial, the recommended phase 2 dose of repotrectinib was 160 mg daily for 14 days, followed by 160 mg twice daily. Response occurred in 56 of the 71 patients (79%; 95% confidence interval [CI], 68 to 88) with ROS1 fusion-positive NSCLC who had not previously received a ROS1 TKI; the median duration of response was 34.1 months (95% CI, 25.6 to could not be estimated), and median progression-free survival was 35.7 months (95% CI, 27.4 to could not be estimated). Response occurred in 21 of the 56 patients (38%; 95% CI, 25 to 52) with ROS1 fusion-positive NSCLC who had previously received one ROS1 TKI and had never received chemotherapy; the median duration of response was 14.8 months (95% CI, 7.6 to could not be estimated), and median progression-free survival was 9.0 months (95% CI, 6.8 to 19.6). Ten of the 17 patients (59%; 95% CI, 33 to 82) with the ROS1 G2032R mutation had a response. A total of 426 patients received the phase 2 dose; the most common treatment-related adverse events were dizziness (in 58% of the patients), dysgeusia (in 50%), and paresthesia (in 30%), and 3% discontinued repotrectinib owing to treatment-related adverse events. CONCLUSIONS: Repotrectinib had durable clinical activity in patients with ROS1 fusion-positive NSCLC, regardless of whether they had previously received a ROS1 TKI. Adverse events were mainly of low grade and compatible with long-term administration. (Funded by Turning Point Therapeutics, a wholly owned subsidiary of Bristol Myers Squibb; TRIDENT-1 ClinicalTrials.gov number, NCT03093116.).

Published
2024

21. Phase III KEYNOTE-789 Study of Pemetrexed and Platinum With or Without Pembrolizumab for Tyrosine Kinase Inhibitor[FIGURE DASH]Resistant, EGFR–Mutant, Metastatic Nonsquamous Non–Small Cell Lung Cancer

Author

Yang, James Chih-Hsin, Lee, Dae Ho, Lee, Jong-Seok, Fan, Yun, de Marinis, Filippo, Iwama, Eiji, Inoue, Takako, Rodríguez-Cid, Jerónimo, Zhang, Li, Yang, Cheng-Ta, de la Mora Jimenez, Emmanuel, Zhou, Jianying, Pérol, Maurice, Lee, Ki Hyeong, Vicente, David, Ichihara, Eiki, Riely, Gregory J., Luo, Yiwen, Chirovsky, Diana, Pietanza, M. Catherine, Bhagwati, Niyati, Lu, Shun, Joseph Boyer, Michael, Hui, Rina, Wong, Mark, Mant, Andrew, Parente, Phillip, John, Thomas, Parakh, Sagun, Castro, Gilberto, Jr, Werutsky, Gustavo, de Azevedo, Sergio Jobim, Andre Franke, Fabio, De Almeida Rego, Joilda Batista, Martins De Marchi, Pedro Rafael, Dix Junqueira, Gustavo, Maris Peria, Fernanda, Brust, Leandro, Cheema, Parneet, Doherty, Mark, Parmar, Ambika, Menjak, Ines B., Leighl, Natasha B., Agulnik, Jason, Lu, Shun, Han, Zhigang, Cui, Jiuwei, Zhang, Li, Cheng, Ying, Chen, Gongyan, Zhang, Helong, Yao, Yu, Hu, Chengping, Wang, Qiming, Zhang, Xin, Zhang, Yong, Zhou, Jianying, Ying, Kejing, Fan, Yun, Wang, Yan, Wang, Ziping, Feng, Jifeng, Du, YingYing, Wu, Lin, Huang, Cheng, Zhou, Xiangdong, Perol, Maurice, Domont, Julien, Lamour, Corinne, Dutilh, Julien, Oulkhouir, Youssef, Westeel, Virginie, Carmier, Delphine, Coudert, Bruno, Lagrange, Aurelie, Spaeth, Dominique, Ropert, Stanislas, Christoph, Daniel C., Kern, Jens, Kopp, Hans-Georg, Griesinger, Frank, Wiewrodt, Rainer, Wermke, Martin, Wesseler, Claas, Mueller, Annette, Vogel, Gunther, Lee, Victor, James Ho, Chung Man, Oscar Chan, Siu Hong, Hung Lo, Sing, Jonathan Nyaw, Shi Feng, Jacky Li, Yu Chung, Bar, Jair, Gottfried, Maya, Dudnik, Julia, Zer, Alona, Moskovitz, Mor, Wollner, Mirjana, Rotem, Ofer, Shamai, Sivan, Asna, Noam, Kamel, Mhameed, Novello, Silvia, Di Costanzo, Francesco, Doni, Laura, Mazzoni, Francesca, Ferrau, Francesco, de Marinis, Filippo, Tonini, Giuseppe, Galetta, Domenico, Piantedosi, Francovito, Vitiello, Fabiana, Kirita, Keisuke, Yoh, Kiyotaka, Takahashi, Toshiaki, Ohe, Yuichiro, Hattori, Yoshihiro, Okamoto, Isamu, Kurata, Takayasu, Yoshioka, Hiroshige, Saka, Hideo, Oki, Masahide, Kato, Terufumi, Tanaka, Hiroshi, Kumagai, Toru, Inoue, Takako, Hida, Toyoaki, Horio, Yoshitsugu, Teraoka, Shunsuke, Ichihara, Eiki, Kishi, Kazuma, Takaya, Hisashi, Takai, Daiya, Kozuki, Toshiyuki, Kasahara, Kazuo, Tambo, Yuichi, Hosomi, Yukio, Kondo, Masashi, Ichiki, Masao, Takeoka, Hiroaki, de la Mora Jimenez, Emmanuel, Hernandez, Carlos Alberto, Rodriguez Cid, Jeronimo Rafael, Rodriguez, Oscar Gerardo Arrieta, Han, Ji-Youn, Min, Young Joo, Kim, Dong-Wan, Park, Keunchil, Lee, Se-Hoon, Lee, Ki Hyeong, Lee, Jong-Seok, Kang, Jin Hyoung, Lee, Dae Ho, Cho, Eun Kyung, Carcereny, Enric, Lopez, Pilar Garrido, Tarruella, Margarita Majem, Rodriguez, Luis Paz-Ares, Vicente Baz, David, Font, Enriqueta Felip, Cobo-Dols, Manuel, Ekman, Simon, Bergman, Bengt, Ohman, Ronny, Vikstrom, Anders, Yang, Chih-Hsin, Chiu, Chao-Hua, Huang, Hsu-Ching, Yang, Cheng-Ta, Su, Jian, Chang, Gee-Chen, Yang, Tsung-Ying, Hsia, Te-Chun, Su, Wu-Chou, Wu, Shang-Yin, Wang, Chin-Chou, Lee, Kang-Yun, Lin, Sheng-Hao, Lin, Chih-Bin, Lee, Jih-Hsiang, Huang, Chun-Yao, Ahmed, Samreen, Newsom-Davis, Thomas, Baijal, Shobhit, Brock, Juliet, Zaki, Kam, Shamash, Jonathan, Papadatos-Pastos, Dionysis, Jain, Pooja, Jane Mackean, Melanie, Kendall, Stephan DiSean, Anderson, Ian, Costin, Dan, Hall, Richard, Campbell, Nicholas, Khan, Saad, Dowell, Jonathan, Mashru, Sandeep, Menon, Smitha, Raza, Ahmad, Ge, Li, Riely, Gregory J., Seetharamu, Nagashree, Stampleman, Laura, Subramanian, Janakiraman, Wender, Donald B., Natale, Ronald B., Zhu, Viola, Ignatius Ou, Sai-Hong, Sanborn, Rachel, and Evangelist, Makenzi C.

Published
2024
Full Text
View/download PDF

22. PA&DA: Jointly Sampling PAth and DAta for Consistent NAS

Author

Lu, Shun, Hu, Yu, Yang, Longxing, Sun, Zihao, Mei, Jilin, Tan, Jianchao, and Song, Chengru

Subjects
Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning
Abstract

Based on the weight-sharing mechanism, one-shot NAS methods train a supernet and then inherit the pre-trained weights to evaluate sub-models, largely reducing the search cost. However, several works have pointed out that the shared weights suffer from different gradient descent directions during training. And we further find that large gradient variance occurs during supernet training, which degrades the supernet ranking consistency. To mitigate this issue, we propose to explicitly minimize the gradient variance of the supernet training by jointly optimizing the sampling distributions of PAth and DAta (PA&DA). We theoretically derive the relationship between the gradient variance and the sampling distributions, and reveal that the optimal sampling probability is proportional to the normalized gradient norm of path and training data. Hence, we use the normalized gradient norm as the importance indicator for path and training data, and adopt an importance sampling strategy for the supernet training. Our method only requires negligible computation cost for optimizing the sampling distributions of path and data, but achieves lower gradient variance during supernet training and better generalization performance for the supernet, resulting in a more consistent NAS. We conduct comprehensive comparisons with other improved approaches in various search spaces. Results show that our method surpasses others with more reliable ranking performance and higher accuracy of searched architectures, showing the effectiveness of our method. Code is available at https://github.com/ShunLu91/PA-DA., Comment: To appear in CVPR 2023; we will update the camera-ready version soon

Published
2023

28. First-line zorifertinib for EGFR-mutant non-small cell lung cancer with central nervous system metastases: The phase 3 EVEREST trial

Author

Zhou, Qing, Yu, Yan, Xing, Ligang, Cheng, Ying, Wang, Ying, Pan, Yueyin, Fan, Yun, Shi, Jianhua, Zhang, Guojun, Cui, Jiuwei, Zhou, Jianying, Song, Yong, Zhuang, Wu, Ma, Zhiyong, Hu, Yanping, Li, Gaofeng, Dong, Xiaorong, Feng, Jifeng, Lu, Shun, Wu, Jingxun, Li, Juan, Zhang, Longzhen, Wang, Dong, Xu, Xinhua, Yang, Tsung-Ying, Yang, Nong, Guo, Yubiao, Zhao, Jun, Yao, Yu, Zhong, Diansheng, Xia, Bing, Yang, Cheng-Ta, Zhu, Bo, Sun, Ping, Shim, Byoung Yong, Chen, Yuan, Wang, Zhen, Ahn, Myung-Ju, Wang, Jie, and Wu, Yi-Long

Published
2025
Full Text
View/download PDF

30. Long-Term Survival Outcomes With First-Line Nivolumab Plus Ipilimumab–Based Treatment in Patients With Metastatic NSCLC and Tumor Programmed Death-Ligand 1 Lower Than 1%: A Pooled Analysis

Author

Peters, Solange, Paz-Ares, Luis G., Reck, Martin, Carbone, David P., Brahmer, Julie R., Borghaei, Hossein, Lu, Shun, O’Byrne, Kenneth J., John, Thomas, Ciuleanu, Tudor-Eliade, Schenker, Michael, Bernabe Caro, Reyes, Nishio, Makoto, Cobo, Manuel, Lee, Jong-Seok, Zurawski, Bogdan, Pluzanski, Adam, Aoyama, Takekazu, Tschaika, Marina, Devas, Vipul, Grootendorst, Diederik J., and Ramalingam, Suresh S.

Published
2025
Full Text
View/download PDF

35. Adeno-associated virus-based gene therapy treats inflammatory kidney disease in mice

Author

Wu, Guochao, Liu, Shuya, Hagenstein, Julia, Alawi, Malik, Hengel, Felicitas E., Schaper, Melanie, Akyuz, Nuray, Liao, Zhouning, Wanner, Nicola, Tomas, Nicola M., Failla, Antonio Virgilio, Dierlamm, Judith, Korbelin, Jakob, Lu, Shun, and Huber, Tobias B.

Subjects
Care and treatment, Physiological aspects, Usage, Health aspects, Gene therapy -- Physiological aspects, Inflammation -- Physiological aspects, Medical research, Kidney diseases -- Physiological aspects -- Care and treatment, Dependoviruses -- Usage -- Health aspects, Medicine, Experimental
Abstract

Introduction Most kidney diseases are associated with the dysfunction of the glomerular filtration barrier (GFB), which comprises 3 layers, including the glomerular endothelial cells (GEC), the glomerular basement membrane (GBM) [...], Adeno-associated virus (AAV) is a promising in vivo gene delivery platform showing advantages in delivering therapeutic molecules to difficult or undruggable cells. However, natural AAV serotypes have insufficient transduction specificity and efficiency in kidney cells. Here, we developed an evolution-directed selection protocol for renal glomeruli and identified what we believe to be a new vector termed AAV2-GEC that specifically and efficiently targets the glomerular endothelial cells (GEC) after systemic administration and maintains robust GEC tropism in healthy and diseased rodents. AAV2-GEC-mediated delivery of IdeS, a bacterial antibody-cleaving proteinase, provided sustained clearance of kidney-bound antibodies and successfully treated antiglomerular basement membrane glomerulonephritis in mice. Taken together, this study showcases the potential of AAV as a gene delivery platform for challenging cell types. The development of AAV2-GEC and its successful application in the treatment of antibody-mediated kidney disease represents a significant step forward and opens up promising avenues for kidney medicine.

Published
2024
Full Text
View/download PDF

36. Bionic driving method of multi-actuator hydraulic synchronization system based on wild goose cluster flight principle

Author

Wang Weiping, Zhou Xinyi, and Lu Shun

Subjects
Mechanical engineering and machinery, TJ1-1570
Abstract

Heavy equipment is often driven synchronously by multi-actuator hydraulic system. However, when the load deviation of the actuators is large, such systems exhibit serious energy loss and inferior synchronization performance. Inspired by the cluster flight of wild goose, a bionic hydraulic actuator and a bionic driving method are proposed. Different from traditional two-chamber hydraulic actuator, the redundant freedom of the bionic hydraulic actuator is used to simulate the airflow vortex formed by wild goose. By doing so, the bionic actuators can provide or accept auxiliary driving force from other actuators according to different topology. In addition, the optimal load balance performance of multi-actuators can be realized by adjusting the auxiliary driving pressure between actuators. In this study, topology switching rules and auxiliary pressure control strategies are discussed. Numerical simulation results demonstrate that the proposed method can reduce the maximum load deviation by over 85% and the synchronization error by over 80% in typical cases.

Published
2024
Full Text
View/download PDF

38. Savolitinib in patients in China with locally advanced or metastatic treatment-naive non-small-cell lung cancer harbouring MET exon 14 skipping mutations: results from a single-arm, multicohort, multicentre, open-label, phase 3b confirmatory study

Author

Yu, Yongfeng, Guo, Qisen, Zhang, Yongchang, Fang, Jian, Zhong, Diansheng, Liu, Baogang, Pan, Pinhua, Lv, Dongqing, Wu, Lin, Zhao, Yanqiu, Li, Juan, Liu, Zhihua, Liu, Chunling, Su, Haichuan, Fan, Yun, Zhang, Tongmei, Liu, Anwen, Jin, Bo, Wang, Ye, Zhou, Jianying, Zhang, Zhihong, Ran, Fengming, Song, Xia, Wang, Mengzhao, Zhang, Xiaodong, Ma, Rui, Song, Zizheng, Li, Xingya, Cheng, Ying, Wang, Changli, Han, Liang, Shi, Jianhua, Yu, Zhuang, Chen, Yuan, Huang, Jianan, Min, Xuhong, Wang, Limin, Lu, Kaihua, Yang, Minglei, Wang, Pingli, Dong, Xiaorong, Yu, Guohua, Wang, Buhai, Chang, Jianhua, Duan, Jianchun, Cang, Shundong, Song, Quanyi, Yu, Jie, Fan, Songhua, Wang, Linfang, Ren, Yongxin, Luo, Xian, Shi, Michael, Su, Weiguo, and Lu, Shun

Published
2024
Full Text
View/download PDF

45. Five-year outcomes with first-line nivolumab plus ipilimumab with 2 cycles of chemotherapy versus 4 cycles of chemotherapy alone in patients with metastatic non-small cell lung cancer in the randomized CheckMate 9LA trial

Author

Reck, Martin, Ciuleanu, Tudor-Eliade, Schenker, Michael, Bordenave, Stephanie, Cobo, Manuel, Juan-Vidal, Oscar, Reinmuth, Niels, Richardet, Eduardo, Felip, Enriqueta, Menezes, Juliana, Cheng, Ying, Mizutani, Hideaki, Zurawski, Bogdan, Alexandru, Aurelia, Carbone, David P., Lu, Shun, John, Thomas, Aoyama, Takekazu, Grootendorst, Diederik J., Hu, Nan, Eccles, Laura J., and Paz-Ares, Luis G.

Published
2024
Full Text
View/download PDF

48. Befotertinib for patients with pretreated EGFR T790M mutated locally advanced or metastatic NSCLC: Final overall survival results from a phase 2 trial

Author

Lu, Shun, Zhang, Yiping, Zhang, Guojun, Zhou, Jianying, Cang, Shundong, Cheng, Ying, Wu, Gang, Cao, Peiguo, Lv, Dongqing, Jian, Hong, Jin, Xiangming, Chen, Chengshui, Tian, Panwen, Wang, Kai, Jiang, Guanming, Chen, Gongyan, Chen, Qun, Zhao, Hui, Ding, Cuimin, Guo, Renhua, Sun, Guoping, Wang, Bin, Jiang, Liyan, Liu, Zhe, Fang, Jian, Yang, Junquan, Zhuang, Wu, Liu, Yunpeng, Zhang, Jian, Pan, Yueyin, Chen, Jun, Yu, Qitao, Zhao, Min, Cui, Jiuwei, Li, Dianming, Yi, Tienan, Yu, Zhuang, Yang, Yan, Zhang, Yan, Zhi, Xiuyi, Huang, Yunchao, Wu, Rong, Chen, Liangan, Zang, Aimin, Cao, Lejie, Li, Qingshan, Li, Xiaoling, Song, Yong, Wang, Donglin, Zhang, Shucai, Ding, Lieming, Zhang, Ling, Ji, Dong, and Shen, Zhilin

Published
2024
Full Text
View/download PDF

49. SpeechNAS: Towards Better Trade-off between Latency and Accuracy for Large-Scale Speaker Verification

Author

Zhu, Wentao, Kong, Tianlong, Lu, Shun, Li, Jixiang, Zhang, Dawei, Deng, Feng, Wang, Xiaorui, Yang, Sen, and Liu, Ji

Subjects
Computer Science - Sound, Computer Science - Computation and Language, Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning, Electrical Engineering and Systems Science - Audio and Speech Processing
Abstract

Recently, x-vector has been a successful and popular approach for speaker verification, which employs a time delay neural network (TDNN) and statistics pooling to extract speaker characterizing embedding from variable-length utterances. Improvement upon the x-vector has been an active research area, and enormous neural networks have been elaborately designed based on the x-vector, eg, extended TDNN (E-TDNN), factorized TDNN (F-TDNN), and densely connected TDNN (D-TDNN). In this work, we try to identify the optimal architectures from a TDNN based search space employing neural architecture search (NAS), named SpeechNAS. Leveraging the recent advances in the speaker recognition, such as high-order statistics pooling, multi-branch mechanism, D-TDNN and angular additive margin softmax (AAM) loss with a minimum hyper-spherical energy (MHE), SpeechNAS automatically discovers five network architectures, from SpeechNAS-1 to SpeechNAS-5, of various numbers of parameters and GFLOPs on the large-scale text-independent speaker recognition dataset VoxCeleb1. Our derived best neural network achieves an equal error rate (EER) of 1.02% on the standard test set of VoxCeleb1, which surpasses previous TDNN based state-of-the-art approaches by a large margin. Code and trained weights are in https://github.com/wentaozhu/speechnas.git, Comment: 8 pages, 3 figures, 3 tables. Accepted by ASRU2021

Published
2021

Catalog

Books, media, physical & digital resources
See catalog results