【華仁全球講座 (H&J Global Chair)】2025/6/12 13:20-14:20專題演講：COSMOS WORLD FOUNDATION MODELS FOR PHYSICAL AI

 

APSIPA Taiwan Chapter 與 MediaTek-NYCU Research Center將於2025年 6月12日（四）13:20-14:20 共同舉辦一場國際級專題演講，

由華仁全球講座贊助下，特邀 NVIDIA 劉洺堉博士 蒞校演說，本次演講將介紹 Cosmos 系列模型，說明其如何協助開發者加速構建智能化且具物理基礎的 AI 應用，並展示相關實際案例。

 

活動詳情請參閱附檔海報及以下資訊，歡迎大家踴躍參與！

 

Title：Cosmos World Foundation Models for Physical AI
Time：13:20 pm - 14:20 pm, June 12
Location：EC 114 (工程三館114室)
Speaker：Dr. Ming-Yu Liu , Vice President of Research at NVIDIA
Host：Prof. Wen-Hsiao Peng, Department of Computer Science

Abstract:

In this talk, I will introduce Cosmos — a family of world foundation models designed to advance Physical AI. The Cosmos framework consists of three core models: Cosmos-Predict, Cosmos-Transfer, and Cosmos-Reason. Cosmos-Predict is a world model that forecasts future states of the physical environment based on current observations and control inputs. Cosmos-Transfer is a conditional generation model that enhances the realism of simulation outputs, supporting robust data augmentation for Physical AI systems. Cosmos-Reason enables embodied agents to perform structured physical reasoning, bridging perception, prediction, and action. Together, these models provide a powerful foundation for building and scaling Physical AI applications. I will showcase how the Cosmos suite empowers developers and researchers to accelerate the development of intelligent, physically grounded agents.

 

Bio:

Ming-Yu Liu is a Vice President of Research at NVIDIA and a Fellow of IEEE. He leads the Deep Imagination Research group at NVIDIA, which focuses on Generative AI for Physical AI. His research team has helped create several new product categories for NVIDIA, including NVIDIA Cosmos, a developer-first world foundation model platform for Physical AI, NVIDIA Edify, a family of Generative AI models that powers Getty Images and Shutterstock's GenAI services, NVIDIA Canvas [GauGAN], a real-time painting tool that uses GANs to turn simple brushstrokes into photorealistic images, and NVIDIA Maxine [LivePortrait], an AI-first cloud-native video streaming platform. His research group constantly publishes scientific papers in top-tier AI conferences regularly, including NeurIPS, ICLR, ICML, CVPR, ICCV, ECCV, and SIGGRAPH. Several of their papers received prestigious awards.

 

聯絡窗口:

Assistant 李珮如, Pei-Zu Lee
Phone: +886-3-5712121 ext. 54820
Fax: +886-3-5721490, Email: pzlee@nycu.edu.tw
Address: EC 344, Dept. of CS, NYCU Taiwan