​谷歌首席工程师分享从机器学习到深度学习的旅程 | Leap Talk

今年夏天，我们推出 Leap Talk 的第一个系列，由世界级产品的高级主管分享 AI, Cloud, Autonomous Driving 及其它高级技术在他们的产品中的实际应用。讲座有 livestream，现场分享结束后会有 social 活动，欢迎大家报名参与，并将好活动分享给身边朋友。

Statistical models and Machine Learning have been used for multiple years in and around the Tech Industry - like other successful innovations, these have stood the test of time, and have had significant impact on products. However, the success of many of the pre-Deep Learning methods have stagnated even in the face of growing data. Deep Learning has the promise to provide a quantum jump and make previously impossible things much easier.

Despite growing evidence that DNNs work, it is a hard decision for a product team to make a switch from their previously used methods. The journey is complex - infrastructure needs to be updated, the training time increases significantly, serving latency and costs can increase, and the variety of models available make it really hard to know what’s the best fit. There is a human element as well - most experienced engineers are not trained to use DNNs, and the transition feels scary. In this talk, Dr. Mehta will share his journey to transition his team in Google Location Platform from using classic ML models to Deep Models, the tradeoffs they had to make, and what their experience was like one year after the transition.

2017年7月15日 

2-3 PM (PDT) Speaking Session

3-4 PM (PDT) Networking

4500 Great America Pkwy,

Santa Clara, CA, 95054