Abstract

David Banks: This is an overview of the complex ecosystem of modern on-line advertising. It will focus on the statistics behind different kinds of recommender systems, as well as topics related to process monitoring of ad success, clickthrough prediction, and optimal contract fulfillment.

Nathaniel Stevens: On-line advertisers run hundreds of designed experiments today. These experiments can be quite complex multiarmed bandits that gain millions of data points within a few minutes. But there are new challenges that arise from interactions among so many experimental arms, incomplete information about users, and explore-exploit tradeoffs.

Hongxia Yang: Artificial intelligence has reached or surpassed human standards in the perceptual intelligence fields such as "listening, speaking, and seeing", but it is still in its infancy in the field of cognitive intelligence that requires external knowledge, logical reasoning, or domain migration. After long-term exploration and verification of large-scale online businesses such as Taobao and Alipay, we have built a "super brain" through the extremely large scale pre-training model M6, and built "flexible limbs" through the edge-cloud collaboration platform Gemini, to fully present the comprehensive picture of the next generation of AI. The 10 trillion M6 is currently the world’s largest pre-training model, achieving the industry’s ultimate low-carbon efficiency. Compared to GPT-3, M6 uses only 1% of the energy consumption which greatly promotes the development of ubiquitous AI. This year, M6 supported the world's largest shopping festival Double 11 for the first time. Relying on its multi-modal understanding ability, M6 greatly improved search and recommendation accuracy of Taobao and Alipay; with its smooth writing ability, M6 created scripts and copywriting for Tmall virtual streamers; depending on its generated high-definition images, M6 has been on duty in Rhino Intelligent Manufacturing and increased the efficiency of designers by 5 times. We developed Gemini, the industry's first edge-cloud collaborative learning framework with the definition of five paradigms: cloud, edge, cloud-centric, edge-centric, and edge-cloud collaboration learning modes. It not only protects data privacy, but also has outstanding performances in personalized recommendations. This talk we will include details of state-of-the-art recommendation system in practice, M6 and Gemini.