구글 제미나이(Gemini)! 너 잼민이 아냐?
[선행AI기술팀 김석겸] 오늘 소개할 기술은 구글의 제미나이입니다. 제미나이는 구글에서 LLM의 최고는 누구인가, 어떤 모달리티까지 커버할 수 있는가, 요즘 핫한 on-device AI까지 다 먹어버리겠다고 나온 모델입니다. 아래 영상이 이를 잘 보여줍니다.…
LaMDA-Google's Conversational Language Model
[Service Development Team Kim Byung-in] At Google I/O 2021, an event that showcases the latest Google technologies, Android, Web, artificial intelligence, Chrome, and other technologies, services, and platform services were released. Among the many technologies, the hottest topic is LaMDA (Google's language…
Google Lyra - speech compression based on deep generative models
Google Lyra is a new voice compression method based on the generative model. Existing voice compression methods have greatly improved the original sound quality, that is, about 8-16kbps required to obtain transparent quality, so that 3kbps low...
Multilingual translation model and language model trained with data from over 100 countries
Many attempts are being made to expand the language model and translation model, which were previously studied mainly in English, into multiple languages. Google's mT5 is a study that extends the existing T5 (text-to-text transfer transformer) into a multilingual corpus, including a total of 101 languages...
Implementation feasibility with Google MixNet
The convolution commonly used in images is a 3D operation. (KxKxC; K=kernel size, C=number of channels) After applying this by dividing it into multiple 2D operations of KxKx1, depthwise separable convolution that applies convolution with a size of 1x1xC in the channel direction greatly reduces the number of parameters...
Google MixIT AI-Separation of unsupervised learning sound sources
MixIT AI, released by Google, is a technology that obtains a separate sound source from single-channel audio in which multiple sound sources are mixed. It can be viewed as a blind source separation task, and unlike existing technologies, it has the feature of delivering excellent performance with unsupervised(!).
