Compression and Optimization of Neural Networks by Ternarization Methods
A.C. TĂNASĂ, D. Tanasa · Bulletin of the Transilvania University of Brasov Series I Engineering Sciences · 2021
Current deep neural networks are achieving higher and higher performances, but this comes to a great computational cost which creates a difficulty to use them on embedded platforms like smartphones, tables, autonomous cars. Fortunately, this can be addressed using quantization techniques, one of which is the ternarization when only 3 bits are used.