Recently, Ali AI has open sourced the new generation of human-machine dialogue model ESIM. The algorithm model has been proposed for more than two years and has been included in Google, facThe international academic community including ebook has been cited in more than 200 papers, and has won the double champion in the international top dialogue system evaluation competition (DSTC7), raising the world record of human-machine dialogue accuracy to 94.1%.
The ESIM model was originally developed by Chen Qian, a post-90s scientist in the Speech Laboratory of Dharma Academy, and has now become a popular model and common standard in the industry. This R&D team with an average age of 30 announced that it will open source the ESIM model to companies and individuals around the world from now on, share this achievement with global developers, and jointly promote the development of artificial intelligence technology.
In the DSTC 7 competition last year, ESIM swept the NOESIS circuit and stood out from nearly 20 participating teams such as Massachusetts Institute of Technology, Johns Hopkins University, and IBM Research. Won the championship of the two races on the track.
DSTC is the authoritative dialogue system evaluation competition in the academic world. It was initiated by scientists from Microsoft Research and Carnegie Mellon University in 2013. It was held this year to the eighth The NOESIS track examines the human-machine dialogue capabilities of AI, and requires AI to follow a given history of multiple rounds of human-machine dialogue, Choose the correct response from hundreds to tens of thousands of sentences.
The human-computer dialogue system and the cognitive intelligence behind it are the most complex and important technology in human-computer interaction, once described by Bill Gates as "the jewel in the crown of artificial intelligence." To make the machine understand quickly and accuratelyFor human expression, ESIM has installed a "radar" system for AI, giving it the ability to retrieve dialogue history in real time and automatically remove interference information, so that it can give the response that humans expect.
This breakthrough will bring significant changes to application scenarios such as smart customer service, navigation software, and smart speakers. The smart voice ordering machine and subway voice ticket vending machine developed by Alibaba based on the ESIM model are already in Hangzhou, Landed in Shanghai and other places.
This is not the first time Ali open sourced cutting-edge technology. In 2018, Dharma Academy open sourced the new generation of speech recognition model DFSMN, which attracted many researchers to work on this model, and even set the world record for speech recognition again.