Computer science researchers at Wayne State earn IEEE IRI best paper award
A research team from Wayne State University received a best paper award at the IEEE International Conference on Information Reuse and Integration (IRI) for Data Science. The 22nd annual IEEE IRI event was held virtually in August.
Qisheng He, a computer science Ph.D. candidate and graduate research assistant at Wayne State, co-authored a paper with Professors Ming Dong and Loren Schwiebert titled, “Octave Deep Compression: In-Parallel Pruning-Quantization on Different Frequencies.”
“Current artificial intelligence and deep neural network models are facing challenges to be deployed on mobile devices due to their big sizes,” said He. “One usually requires hundreds of megabytes to be stored.”
To overcome this challenge and still achieve accuracy in visual recognition tasks, compression is necessary to store and run the models on a cell phone or on-board vehicle system. Combining deep compression technology and Octave Convolutional Network, the research team developed a novel frequency-wise compression method called Octave Deep Compression (ODC).
According to He, ODC can compress an AI model up to 100 times and reach the same or higher performance in many applications, including visual object recognition. The compressed model only needs several megabytes and can be easily stored and efficiently operated on mobile devices.