Flexible Quantization for Efficient Convolutional Neural Networks

• Autores: Zacchigna, Federico Giordano; Lew, Sergio Eduardo; Lutenberg, Ariel
• Año de publicación: 2024
• Idioma: inglés
• Tipo de recurso: artículo
• Materia: CNN; quantization; uniform; non-uniform; mixed-precision; FPGA; ASIC; edge devices; embedded systems
• Institución: Consejo Nacional de Investigaciones Científicas y Técnicas

This work focuses on the efficient quantization of convolutional neural networks (CNNs). Specifically, we introduce a method called non-uniform uniform quantization (NUUQ), a novel quantization methodology that combines the benefits of non-uniform quantization, such as high compression levels, with the advantages of uniform quantization, which enables an efficient implementation in fixed-point hardware. NUUQ is based on decoupling the quantization levels from the number of bits. This decoupling allows for a trade-off between the spatial and temporal complexity of the implementation, which can be leveraged to further reduce the spatial complexity of the CNN, without a significant performance loss. Additionally, we explore different quantization configurations and address typical use cases. The NUUQ algorithm demonstrates the capability to achieve compression levels equivalent to 2 bits without an accuracy loss and even levels equivalent to ∼1.58 bits, but with a loss in performance of only ∼0.6%.
Fil: Zacchigna, Federico Giordano. Universidad de Buenos Aires. Facultad de Ingeniería. Departamento de Electronica; Argentina
Fil: Lew, Sergio Eduardo. Universidad de Buenos Aires. Facultad de Ingeniería. Departamento de Electronica; Argentina
Fil: Lutenberg, Ariel. Universidad de Buenos Aires. Facultad de Ingeniería. Departamento de Electronica; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas; Argentina