Invidia, Nvidia, Assault discussed on Google Cloud Platform Podcast

Automatic TRANSCRIPT

By this open paper doing unsupervised learning for tex classifications pacific lee sentiment analysis where you train a model to reconstruct a large text database character by character and it turns out that in so doing your model has to understand the meaning of the text not just the characters in the tax and you can use that with just a little bit of labels to to solve some some really applied problems like sentiment analysis and so we were working on stuff like that and then in the audio space right now we're working on some things related to speech and speech synthesis so so in terms of doing this type of work you're trying to also set it up to make it acceptable to users of the gps i assume absolutely yeah one of the great things about working at invidia is that we get to open source a lot of the work that we're doing and in fact all the products that i'm mentioning we have either open source or in the process of open sourcing and one of the reasons that we can do that as because the more are we inspire other people to do than the better it is for invidia because so much of a happens on gp's and so we can make the world a better place by giving people new technology that they can use assault problems and it also helps video business and so it's it's a great thing for us to open source our work right and then i know you've done the work in the past and cudi n n and really driving software to help people often is working on gp's but then there's gp that have been actually built specifically volta to help with deep learning can you speak to a little bit about volta and how it works and helps people improve their processing of hefty learning yeah so volta is the first gp that nvidia made kind of after it woke up to the possibilities of deep learning and invidia believes i think more strongly than than any other organization that i've seen in the power of specialized architecture you know that's kind of really what has made.

Coming up next