Computer Science and Applied Mathematics

Supervised Deep Learning
Marc’Aurelio Ranzato, Facebook

In supervised learning, each input sample is provided with a target label dur¬ing training. In this talk, I will describe how deep learning methods, which are algorithms vaguely inspired by how the brain works, can be trained to predict the label of unseen inputs for three different applications: speech recognition, text understanding and generic object recognition in images. In the first ap¬plication, the input is 100ms of speech and the output is a phone label of the sound, in the second case the input is a sequence of words and the output is the subsequent word, in the last case the input is an image and the output is the label of the category of the object in the image (e.g., “dog”). Although these applications are very different from each other, the learning algorithm is very similar. These methods have yielded the most accurate prediction systems on a variety of tasks, and they have recently been deployed in several commercial systems (e.g., speech recognition on Android phones and image search on Baidu search engine, to name a few).

# vimeo.com/80026009 Uploaded 57 Plays / / 0 Comments Watch in Couch Mode

Follow

Computer Science and Applied Mathematics

Kavli Frontiers of Science Plus

The Kavli Frontiers of Science symposium series is the National Academy of Science’s premiere activity for distinguished young scientists. Unlike meetings that focus on a narrow area of science, these meetings allow participants to explore innovative


+ More

The Kavli Frontiers of Science symposium series is the National Academy of Science’s premiere activity for distinguished young scientists. Unlike meetings that focus on a narrow area of science, these meetings allow participants to explore innovative research ideas across a wide variety of fields and to develop new networks that will serve them as they progress in their careers..

Browse This Channel

Shout Box

Channels are a simple, beautiful way to showcase and watch videos. Browse more Channels. Channels