Date/Time
Date(s) - Mon 16 September
15:00 - 16:00
Location
Cotton Club, Cotton 350, VUW
James Hepi is scheduled to present a seminar about: Native Plant Recognition by Machine Learning on a Mobile Device
School of Engineering and Computer Science PhD Proposal
Abstract:
The use of machine learning systems has advanced and evolved with each new generation of computer hardware, allowing the integration of software applications and implementations that were previously only run on home computer systems or larger systems. Most current machine learning techniques for image recognition are based on a Convolutional Neural Network (CNN). The most common method of computer image analysis is through the use of a convolutional neural network (CNN), which has a large computational overhead requirement. However, with the limited computational resources available in a standard mobile device, the training phase would overtax the mobile device hardware, causing analysis failure or, at worst, hardware failure of the device. However, there are a number of specific limitations in implementing a CNN on a mobile device, such as not being able to fully utilize hardware parallelism, energy limitations, and computational power. This research aims to address these problems by testing various CNN models to provide a working base and then developing new algorithms according to the test results. An alternative or developing type of system would be more advantageous, such as the incremental learning CNN model, as it continually improves image classification accuracy as the modeling data are gathered incrementally over time without forgetting previous learning data. In order to achieve the required result, this project will need to achieve a number of predetermined objectives.
First: Address the challenges of small data sets for NZ plant varieties.
Second: Implement an incremental learning technique. Third: Ensure that any privacy issues are taken care of appropriately.
Any queries about this seminar, please email Tony.McLoughlin@vuw.ac.nz