Running Deep Learning Models on Smartphones as Real-Time Apps for Signal and Image Processing Applications

In many signal and image processing applications, deep learning models or deep neural networks have provided superior performance compared with conventional machine learning solutions. This talk covers how deep learning models can be turned into apps running in real-time on smartphones (both Android and iOS). One signal and one image processing application are presented. The image processing application involves real-time implementation of a deep learning model as a smartphone app to detect retinal abnormalities in an on-the-fly manner as retina images are captured by the smartphone camera through commercially available lenses. The motivation behind this application is to use smartphones as an alternative to fundus cameras providing a cost-effective and widely accessible approach to first-pass eye examination. The signal processing application involves real-time implementation of the speech processing pipeline of hearing aids as a smartphone app. The components of the implemented pipeline include a deep learning-based voice activity detection, noise reduction, noise classification, and compression. The motivation behind this application is to use smartphones as an open-source, programmable, and portable signal processing platform to conduct hearing enhancement studies in realistic audio environments.

Speaker(s): Prof. Nasser Kehtarnavaz, Univ. of Texas at Dallas,

Agenda:

6:00 PM Social; 6:30 PM Dinner; 7:15 PM Presentation; 8:45 PM Wrap-up

Location:
Room: Ask for IEEE room at front counter
Bldg: The Doubletree Club – Orange County Airport
7 Hutton Centre Drive
Santa Ana, California
92707