◆  ALIYUN Industry Talk 1 (Time: Nov.2 10:15-10:30)

Title: Multilingualism of LLMs: From Foundation Model to Applications

Speaker: Baosong Yang, Research Scientist at Alibaba Tongyi Lab

Short Bio: Baosong Yang is currently a Scientist at Alibaba's Tongyi Lab, where he is responsible for multilingual and translation algorithms. He received his Ph.D. from the NLP2CT Lab at the University of Macau. Baosong have studied and practiced in the broad field of multilingualism in natural language processing (NLP), with a special emphasis on large language models and machine translation. He has published over 50 papers in leading NLP and AI journals and conferences such as ACL, EMNLP, AAAI, and NeurIPS.

Abstract: Multilingual and cross-lingual capabilities significantly boost the flexibility and usefulness of large language models (LLMs). In this presentation, using Qwen as an example, we'll explore methods to enhance multilingual performance in LLMs, including training, fine-tuning, and evaluation strategies. Additionally, we'll examine the real-world applications of these advancements through Gummy, an end-to-end speech translation model built on Qwen, demonstrating how multilingual capabilities can create practical solutions that overcome language barriers and promote smooth communication.

◆  OPPO Industry Talk 2 (Time: Nov.2 11:45-12:00)

Title: Lightweight Technology of Large Language Models and Its Application Practices in OPPO

Speaker: Feng Zhou

Short Bio: Feng Zhou is the lead of the lightweight technology and its application direction of language models in the AI Center of OPPO, who has been engaged in natural language processing research and application for more than ten years at Baidu, Microsoft, and OPPO. Currently, mainly responsible for the development of lightweight algorithms for large language models and AI applications on mobile devices at OPPO.

Abstract: Large language models have rich application scenarios on mobile devices. To address the data privacy and security needs of phone users, as well as to improve the experience in low-network conditions, there is a growing demand for the edge deployment of large language models. In view of this, we have constructed lightweight technologies such as pruning, distillation, and quantization to compress large language models with low loss to a scale suitable for deployment on the edge side. Furthermore, we designed the 1+N multi-LoRa solution, implementing only a base model on the edge side to support the multiple application scenarios.

◆  JD Industry Talk 3 (Time: Nov.3 10:00-10:15)

Title: Generative AI：Technical Frontier and Industry-Scale Practices

Speaker: Xiaodong He

Short Bio: Dr. He Xiaodong is a world-class AI scientist, who has received a bachelor degree from Tsinghua University, a MS degree from Chinese Academy of Sciences, and a PhD degree from the University of Missouri-Columbia. Before joining JD.com, Dr. He Xiaodong served as the Principal Researcher and Research Manager of the DLTC at Microsoft Research, Redmond in the US. He has published more than 200 theses related to artificial intelligence areas including natural language processing and multi-modal intelligence on language and vision, which have been cited over 50,000 times by Google Scholar, with his top 10 theses cited over 15,000 times. He has also won a slew of awards such as ACL Outstanding Paper Award and IEEE SPS Young Author Best Paper Award. He once served as the Chair of IEEE Seattle Section, and held an editorial position at several top journals. In 2021, he was included in the AI 2000 World Most Influential Scholar List, which was released by Tsinghua University-Chinese Academy of Engineering Knowledge Intelligence Joint Research Center. He was selected in three areas, including natural language processing, speech recognition, information retrieval and recommendation, making him one of the only 60 scientists across the world who were included in the list. After joining JD.com in 2018, Dr. He Xiaodong has set up a series of intelligent man-machine interaction technology laboratories in Beijing, Chengdu and Silicon Valley, as well as an Intelligent Customer Service Product Department to promote the industrialization of technology. With his team, he has developed the first large-scale commercially applied emotional intelligent customer service system, which has served more than 500 million users as JD Retail’s core system, and achieved many bench-marking cases and successful practices.
Personal Title：IEEE Fellow & CAAI Fellow
Vice President of JD.com
Head of JD Smart Customer Service Product Department
Affiliate Professor at the University of Washington
Fellow of Beijing Academy of Artificial Intelligence

Abstract: The report will introduce the underlying technologies of artificial intelligence, including advancements in AI foundation large language models and multimodal large models. Using JD's Industry Practices as examples, it will discuss disruptive innovations in content production, services, and marketing, particularly breakthroughs in multimodal digital humans and cutting-edge explorations in embodied intelligence. The report not only explains the development trends of fundamental technologies but also combines industry case studies to provide insights and reflections on the large-scale industrial application of AI technologies.