Since the release of ChatGPT at the end of November 2022, the Large Language Model has quickly become the focus of the industry. With the continuous upgrading of the large-scale foundation model, the large language model shows increasingly impressive level of intelligence. The powerful ability of the large language model has opened the prelude to the generalization of artificial intelligence, providing all sectors with the ability to optimize business scenario efficiency, thereby promoting industrial change.

Holistic Artificial Intelligence is about systematically reconstructing artificial intelligence technology around industry applications, computing power, and models. The aim is to realize intelligence in the entire process and in all aspects from AI-related production factors to production objects and business objectives. Relying on ubiquitous networks and AI computing power, intelligent management and operation of all factors and scenarios can be realized, while ensuring the confidence, controllability, and security of AI businesses. With business leading the way, AI technology is deeply integrated into the actual scenarios of social production and life.

Focusing on large language models and Holistic Artificial Intelligence, this workshop explores the model of "Large Language Model + Holistic AI". We have invited well-known scholars and experts to deliver keynote speeches, sharing their explorations and understanding of Holistic AI and large language models, and discussing how to realize the generalized application of "Holistic AI + Large Language Model" in various scenarios.

Time: 13:45-14:15

Title: Holistic AI: Thinking and Practice

Abstract : The speech explores the thought and practice of Holistic AI. The speaker will share the thought about Holistic AI and discuss the relationship between large language models and Holistic AI, such as the method of construction of autonomous and controllable general foundation large models suitable for universal scenarios. Also, the speaker will discuss the practice of Holistic AI in various scenarios，including network, customer service, government affairs, medical and other industries. In addition, the speaker will discuss the possibility of construction of high-reliability, high-dynamic self-adaptive industry large language models.

Speaker: Junlan Feng, IEEE Fellow, Chief Scientist of China Mobile, Vice Chairman of the China Artificial Intelligence Industry Alliance, Board Chair of Linux Foundation Network (2020-2023). Dr. Feng received her Ph.D. on Speech Recognition from Chinese Academy of Sciences, and joined AT&T Labs Research in 2001, as a principal researcher until 2013. Dr. Feng joined China Mobile Research in 2013. Since then, she has been leading the AI R&D center of China Mobile - Jiutian. She and the Jiutian team have published over 150 technical research papers, hold 700+ patents, won 16 AI algorithm competition awards, won 20+ national-level and province-level technical awards, undertaken 16 national-level research projects, initiated and led 86% international standards on network intelligence. Jiutian products and AI models under her leadership have been deployed to 3700+ production services and contributed yearly commercial value of 3.9 billion Chinese Yuan. Dr. Feng is a frequent reviewer and program member of major top AI conferences and journals. She had been elected as a member of IEEE speech and natural language processing committee, IEEE industry committee, etc.

Time: 14:20-14:50

Title: Knowledge-Enhanced Large Language Models: Knowledge Injection and Retrieval Enhancement

Abstract : In this talk, Xiaocheng Feng will provide a comprehensive insight into knowledge-enhanced large language models for both academic and industrial researchers. Regarding how to use external knowledge bases and corpus to enhance large language models, he will introduce two methods: knowledge injection and retrieval enhancement. Then, he will discuss their potential and limitations in solving real-world problems. In addition, he will introduce his recent work: retrieval-generation synergy to augment large language models. Finally, he will discuss current research challenges and potential future directions. Finally, he will discuss current research challenges and potential future directions.

Speaker: Xiaocheng Feng ,Associate Professor of the Faculty of Computer, Harbin Institute of Technology (HIT). He is a member of the Research Center for Social Computing and Information Retrieval (SCIR) and leads the Text Generation team. His research topics currently focus on text generation and machine translation. Dr. Feng achieved the China National Conference on Computational Linguistics (CCL) 2021 BEST Paper Award. He was selected as a member of the Young Elite Scientist Sponsorship Program by the China Association for Science and Technology, 2020-2022, and his Ph.D. Dissertation was selected as one of the Best Ph.D. Dissertations by the Chinese Information Processing Society of China.

Time: 14:55-15:25

Title: Method for Aspect-based Sentiment Analysis on Social Comment Text

Abstract : In this talk, we will introduce a new framework designed for accurately aspect sentiment triplet extraction. By introducing rich syntactics and flexible semantics and interactively fuse them to encode into the text representation, we try to learn more task-related information and improve the model's ability to extract aspect sentiment triplets. Furthermore, we also focusing on developing methods for cross domain aspect sentiment triplet extraction. We employ generative model with prompt-tuning to generate target domain data and the pseudo-labels to achieve the goal of triplet extraction from the unlabeled target domain.

Speaker: Tao Qin, Full Professor of the department of Computer Science and Technology, Xi’an Jiaotong University, China. His current research interests include online social network analysis, network traffic analysis and network security. His research achievements have won the 2008 IEEE CSIM Best Paper Award, the First Prize of Science and Technology Progress in Shaanxi Province in 2013 and 2023 respectively, and the Best Resource Paper Award in CCKS 2021.

Time: 15:30-16:00

Title: Knowledge-Retrieval Dialog Systems with Semi-Supervision

Abstract : Recently, a progress in question answering and document-grounded dialog systems is retrieval-augmented methods with a knowledge retriever. Inspired by such progress, we propose a retrieval-based method to enhance knowledge selection in task-oriented dialog systems, which significantly outperforms the traditional database query method for real-life dialogs. Further, we develop latent variable model based semi-supervised learning, which can work with the knowledge retriever to leverage both labeled and unlabeled dialog data. Joint Stochastic Approximation (JSA) algorithm is employed for semi-supervised model training, and the whole system is referred to as JSA-KRTOD. Experiments are conducted on a real-life dataset from China Mobile Custom-Service, called MobileCS, and show that JSA-KRTOD achieves superior performances in both labeled-only and semi-supervised settings.

Speaker: Zhijian Ou, Associate Professor of the Department of Electronic Engineering, Tsinghua University. He received his Ph.D. from Tsinghua University in 2003. He currently serves as Associate Editor of “IEEE/ACM Transactions on Audio, Speech and Language Processing”, Editorial Board Member of “Computer Speech and Language”, member of IEEE Speech and Language Processing Technical Committee, and was General Chair of SLT 2021, Tutorial Chair of INTERSPEECH 2020, General Chair of EMNLP 2022 SereTOD Workshop. He has actively led national research projects as well as research funding from China Mobile, He has actively led national research projects as well as research funding from China Mobile, China Unicom, Apple, IBM, Intel, Panasonic, Toshiba and so on. His research interests are AI speech and language processing.

Time: 16:05-16:35

Title: Network Intelligence with the Wave of Big Models

Abstract : The speech explores how to implement network intelligence in the wave of large models. First, the author discusses the relationship between large language models and network intelligence, and analyze the methods for implementing network intelligence from the perspective of tasks, data, and intelligent models. Finally, the speaker shares the challenges and emerging problems faced in implementing systematic network intelligence.

Speaker: Yong Zhang, Professor of the department of Electronic Engineering, Beijing University of Posts and Telecommunications. He was the Director of Fab.X Artificial Intelligence Research Center, BUPT. He is the Deputy Head of the mobile internet service and platform working group, China communications standards association. In the past five years, he has been working in the field of network intelligence, including network traffic prediction, network anomaly detection, root cause analysis, intelligent resource allocation, the integration of intelligence service, computing and network, etc.