Speakers（报告人）

Zhaoxiang Zhang (Institute of   Automation, Chinese Academy of Sciences, China)

Yi Yang (Zhejiang University, China)

Shuqiang Jiang (Institute of Computing   Technology, Chinese Academy of Sciences, China)

Cewu Lu (Shanghai Jiao Tong University,   China)

Speaker 1

Speaker: Zhaoxiang Zhang (Institute of Automation, Chinese Academy of Sciences,   China)

Title: World Simulator: Exploration and   Fusion of Multipath

Abstract:The development of artificial intelligence   is evolving rapidly. On one hand, new technologies represented by multimodal   large models and generative large models are emerging constantly; on the   other hand, new applications represented by embodied intelligence and Agents   are continuously deepening. Among the integration of these technologies and   applications, world simulators are the most crucial core enabling technology.   This   report delves into the significant value and feasibility of world simulators,   exploring key technical approaches and our preliminary findings. Practical   applications in areas such as intelligent perception, autonomous driving,   robotics, internet control, and smart cities are discussed to illustrate   their effectiveness. Finally, the report outlines the prospects for the   multi-path integration of world simulators.

Biography: Zhaoxiang Zhang, Ph.D., is a   researcher, doctoral supervisor, Changjiang Distinguished Professor, and   executive deputy director of the New Laboratory of Pattern Recognition at the   Institute of Automation, Chinese Academy of Sciences, and a professor at the   University of Chinese Academy of Sciences. His research interests include   pattern recognition, embodied intelligence, and agent learning. He has published over   200 papers in top journals such as IEEE T-PAMI, IJCV, JMLR, National Science   Review and top conferences including CVPR, ICCV, ECCV, NeurIPS, ICLR, AAAI,   IJCAI. He was granted over 30 patents. He has led or is leading multiple   national research projects including the National Natural Science Foundation   of China Key Project, Key International (Regional) Cooperation Research   Project, Joint Key Support Project with CETC, National Key R&D Project. He   served as Area Chair for top conferences like CVPR, ICCV, NeurIPS, ICLR   multiple times. He won the First Prize for Scientific and Technological   Progress of Beijing Science and Technology Award as the first accomplisher.

Speaker 2

Speaker: Yi Yang (Zhejiang University, China)

Title: Content Generation and Engineering Simulation   based on Knowledge-Driven Artificial Intelligence

Abstract: Knowledge-driven artificial intelligence for domain-specific   applications focuses on efficiently integrating pre-trained large models,   prior knowledge, and domain-specific models. This talk will cover   multi-knowledge-driven content generation technologies for applications such   as digital human reconstruction, animation, and cross-media content   generation. First, the talk will explore methods that incorporate geometric   and other prior information for digital human reconstruction and animation.   Next, it will discuss techniques for controllable content generation through   audio, text, layout structure, and other data. Then, practical case studies   will be presented to highlight hybrid model collaboration mechanisms, such as   specialized knowledge embedding and structured expression. Finally, the talk   will introduce the use of generative AI in engineering simulation.

Biography:Yi Yang   is a Qiushi Chair Professor at Zhejiang University. He currently serves as   the Vice Dean of the College of Computer Science at Zhejiang University, and   Director of the Microsoft-Ministry of Education Key Laboratory of Visual   Perception. His research focuses on artificial intelligence and its   applications. His published papers have received over 70,000 citations on   Google Scholar, with an H-index of 131. He has been consecutively named a   Clarivate Analytics Highly Cited Researcher for the past 6 years. He has   received numerous international awards in the AI field, including the   National Outstanding Doctoral Dissertation Award from the Ministry of   Education (2010), the Australian Research Council DECRA Fellowship (2013),   the Australian Computer Society Gold Disruptor Award (2016), Google Faculty   Research Award (2016), Australian Career Achievement Award (2019), Amazon   Machine Learning Research Award (2020), AAAI Most Influential Paper Award   (2021), and the Best Paper Award at ACM MM (2023). He has also led teams to   win over 20 world championships in international research competitions.

Speaker 3

Speaker: Shuqiang Jiang   (Institute of Computing Technology, Chinese Academy of Sciences, China)

Title: Embodied Navigation Combining Exploration and Imagination.

Abstract:Embodied AI represents a significant manifestation of   artificial intelligence in the real physical world, which has showcased great   application potentials in dynamic open-world environments. Embodied   navigation refers to the ability of the agent to perceive and understand the   environment based on task objectives (such as language instructions), then   predict and execute movement actions, thereby progressively completing tasks.   It is the key technology for embodied intelligent systems to interact with   the real world. Existing methods for embodied navigation largely rely on   current and past visual observations for short-term and single-step action   prediction, lacking the capability for evaluating unobserved environments and   conducting long-term action planning. Physiological studies have indicated   that humans not only depend on current observations but can also imagine   unobserved environments from prior memories, constantly refining and   enhancing their understanding of the environment by combining exploration and   imagination. Thus, endowing agents with the ability to “imagine” thereby   aiding them in predicting the layout of unobserved environments, assessing   the long-term value of navigation actions, and realizing more efficient and   accurate navigation decisions, emerges as a significant research challenge.   This report will first introduce the research background of embodied AI and   embodied navigation and then report on the research progress in embodied   navigation combining exploration and imagination, including self-supervised   generative map and lookahead exploration with neural radiance representation,   and finally introduce the adaptation of embodied navigation from simulator to   the real world, providing demonstrations.

Biography: Shuqiang Jiang   (SM’08) is a professor at the Institute of Computing Technology (ICT),   Chinese Academy of Sciences (CAS) and a professor at the University of CAS.   He is also affiliated with the Key Laboratory of Intelligent Information Processing, CAS. His   research interests include multimedia analysis and multimodal intelligence.   He has authored or coauthored more than 200 papers on the related research   topics. He was supported by National Science Fund for Distinguished Young   Scholars in 2021. He won the CAS International Cooperation Award for Young   Scientists, the CCF Award of Science and Technology, Wu Wenjun Natural   Science Award for Artificial Intelligence, CSIG Natural Science Award, and   Beijing Science and Technology Progress Award. He is the Associate Editor of   IEEE TMM and ACM ToMM, vice Chair of IEEE CASS Beijing Chapter, vice Chair of   ACM SIGMM China chapter. He has served as an organization member of more than   20 academic conferences, including the general chair of ICIMCS 2015, program   chair of ICIMCS2010, PCM2017, ACM Multimedia Asia2019, He has also served as   an area chair or TPC member for many conferences, including ACM Multimedia,   CVPR, ICCV, IJCAI, ICME, ICIP, etc.

Speaker 4

Speaker: Cewu Lu   (Shanghai Jiao Tong University, China)

Title: Exploring the Embodied Intelligence PIE Framework: Perception(P),   Imagination(I), Execution(E), and the Role of Large Models

Abstract: This lecture introduces the speaker’s Embodied Intelligence PIE   solution. P (Perception) introduces the full perception and interactive   perception of robots. I (Imagination), an introduction to the speaker’s   conceptually driven simulation reasoning framework for the physical world. E   (Execution) introduces the ideas and execution of the general meta-operation   skills. Based on the above three modules, the exploration and preliminary   results of the embodied PIE large model are introduced. Finally, the work of   embodied cognitive intelligence is introduced, focusing on verifying the stable implicit relationship   between brain neural behavior and physical behavior.

Biography:Lu Cewu   is a professor and Changjiang Distinguished Professor of Shanghai Jiao Tong   University. He is the winner of the Xplorer Award, and was awarded the   Overseas High-level Youth Talent Introduction in 2016. In 2018, he was   selected as 35 Innovators Under 35 (MIT TR35) by MIT Technology Review. In   2019, he was awarded Qiu Shi Outstanding Young Scholar. In 2020, he was   awarded the Special Prize of Shanghai Science and Technology Progress Award   (ranked third). In 2022, he won the Outstanding Scientific Research   Achievement Award of Colleges and Universities from the Ministry of   Education, and got one of the Best Papers Award in IROS (6/3579). In 2023, he   was nominated for the Best Paper in RSS (four in total). He has published   more than 100 papers in high-level journals and conferences such as Nature,   Nature Machine Intelligence, TPAMI as the corresponding author or the first   author. He is a reviewer for Sicence, Nature, Cell and other journals, and   the area chair of NeurIPS, CVPR, ICCV, ECCV, IROS, ICRA. His Research   interests include embodied intelligence and computer vision.