DBTB INT Shubho Sengupta r

yeah I'm shuba shuba sengupta I work for baidu at Silicon Valley AI lab as part of the research group I'm just coming here and sharing about the way we train our networks and especially training networks with really big data which I think they're the forefront of doing it and I'm sharing that knowledge with well without data we wouldn't even be able to train on networks that networks are really large many hundreds of millions of parameters and the reason the work is we have I think saw the largest and best data sets for training them so and data is our essence in some sense without it just won't work and the main insight that I want people in from is if you are coming into a field with a little experience and you have a lot of data then you can use the power of big data and the ability to train big networks to go up to speed really really fast and we have seen this with deep learning over and over again that very quickly deep learning is given our state of the art performance at bar with humans sometimes even better than new ones it all works because we have we know now how to train these networks on the large data set all i can say is just practice practice practice train a lot of networks is a very very experimental field it's not a lot of theory especially if you want to go into deep learning I train models every day to start ideas every day and learn from the data every day so go to goggle or if you're on your own just run a lot of experiments in to store your ideas