Jianfeng Gao的肖像

Jianfeng Gao

Technical Fellow & Corporate Vice President

新闻与深度文章

animation showing hyperparameters
微软研究院博客

| Edward Hu, Greg Yang, 和 Jianfeng Gao

Great scientific achievements cannot be …

Clues diagram
微软研究院博客

| Jianfeng Gao 和 Ahmed Awadallah

Recent progress in natural language unde…

SuperGLUE leaderboards showing T-NLRv5 at the top
微软研究院博客

| Jianfeng Gao 和 Saurabh Tiwary

As part of Microsoft AI at Scale (opens …

Diagram: The proposed SOLOIST model architecture and training objectives. Dialog history, Belief state, DB state, and Response make up the pipeline. Task 1, belief state prediction, corresponds with belief state. Task 2 and Task 3, grounded response generation and contrastive objective correspond with response. A user is shown thinking a goal, which points from the dialog history to the user (response), and then back to dialog history (input). Belief state points down to an image of a computer server (Belief state query) and then back to DB State (DB state results). The server points to readouts labeled “entity.”
微软研究院博客

| Baolin Peng, Chunyuan Li, Jinchao Li, Lars Liden, 和 Jianfeng Gao

The increasing use of personal assistant…

graphical user interface, text, application, email
微软研究院博客

| Rangan Majumder, Alec Berntson, Daxin Jiang (姜大昕), Jianfeng Gao, Furu Wei, 和 Nan Duan

Azure Cognitive Search (opens in new tab…

a graphic shows computer vision model using example of dog image
微软研究院博客

| Chunyuan Li, Lei Zhang, 和 Jianfeng Gao

Humans perceive the world through observ…

微软研究院博客

| Pengchuan Zhang, Lei Zhang, 和 Jianfeng Gao

Humans understand the world by perceivin…

微软研究院博客

| Pengcheng He, Xiaodong Liu, Jianfeng Gao, 和 Weizhu Chen

Natural language understanding (NLU) is …

a screenshot of a cell phone
微软研究院博客

| Hoifung Poon 和 Jianfeng Gao

COVID-19 highlights a perennial problem …