Automatic Approach to Design Efficient Deep Neural Networks

Deep neural networks have demonstrated state-of-the-art modeling accuracy on a wide range of real-life problems, with some cases surpassing human performance. Despite the promise of deep neural networks as an enabling technology for a large number of industries and fields, there are two particular key challenges in the design of deep neural networks in real-world, operational scenarios. First, the design of deep neural networks is a very time consuming process for a machine learning expert, and often results in complex, non-optimal deep neural networks. Secondly, many of the deep neural networks designed in recent years, while achieving high accuracy, are often very complex with high millions to billions of parameters, making such networks intractable for real-time scenarios on mobile device, IoT devices, and other embedded devices. TO BE CONT’D

Faculty Supervisor:

David Clausi

Student:

Seyed Mahmoud Famouri

Partner:

DarwinAI

Discipline:

Engineering - other

Sector:

Information and communications technologies

University:

Program: