Presented By
O’Reilly + Intel AI
Put AI to Work
April 15-18, 2019
New York, NY

Automated neural architecture search for efficient deep learning

Bichen Wu (UC Berkeley)
4:05pm4:45pm Wednesday, April 17, 2019
Machine Learning, Models and Methods
Location: Grand Ballroom West
Secondary topics:  Computer Vision, Deep Learning and Machine Learning tools

Who is this presentation for?

AI product developer & manager, AI researchers, CTOs

Level

Intermediate

Prerequisite knowledge

Basic understanding of deep learning, computer vision.

What you'll learn

- Manually designing neural networks is slow and the results are sub-optimal. - Automated neural architecture search can help us discover more accurate and efficient models, greatly accelerating the development cycle of AI products.

Description

Many computer vision applications rely on accurate and efficient neural networks. For years, we’ve been manually designing and optimizing neural networks to meet the accuracy and efficiency requirements. However, the productivity of such manual design flow is extremely low — It can take months to develop one model and optimize it for one target computing platform. Limited by productivity, we can only afford to design one neural network but deploy it on many computing platforms, even though this leads to sub-optimal performance. To solve this problem, we propose a scalable, easy-to-use, and automated framework for designing & optimizing neural networks. With small computing costs, this framework automatically designs highly accurate models customized for each computing platform with superior efficiency.

Photo of Bichen Wu

Bichen Wu

UC Berkeley

Bichen Wu is a PhD student at UC Berkeley, where he focuses on deep learning, computer vision, and autonomous driving.

Leave a Comment or Question

Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?

Join the conversation here (requires login)