Timezone: »

FleXOR: Trainable Fractional Quantization
Dongsoo Lee · Se Jung Kwon · Byeongwook Kim · Yongkweon Jeon · Baeseong Park · Jeongin Yun

Wed Dec 09 09:00 AM -- 11:00 AM (PST) @ Poster Session 3 #846
Quantization based on the binary codes is gaining attention because each quantized bit can be directly utilized for computations without dequantization using look-up tables. Previous attempts, however, only allow for integer numbers of quantization bits, which ends up restricting the search space for compression ratio and accuracy. In this paper, we propose an encryption algorithm/architecture to compress quantized weights so as to achieve fractional numbers of bits per weight. Decryption during inference is implemented by digital XOR-gate networks added into the neural network model while XOR gates are described by utilizing $\tanh(x)$ for backward propagation to enable gradient calculations. We perform experiments using MNIST, CIFAR-10, and ImageNet to show that inserting XOR gates learns quantization/encrypted bit decisions through training and obtains high accuracy even for fractional sub 1-bit weights. As a result, our proposed method yields smaller size and higher model accuracy compared to binary neural networks.

Author Information

Dongsoo Lee (Samsung Research)
Se Jung Kwon (Samsung Research)

Se Jung Kwon received the Ph.D. degree with specialization in systems modeling and simulation from the Department of Electrical Engineering, KAIST, Daejeon, South Korea, in 2018. He is currently a staff engineer for on-device AI at Samsung Research, Seoul, South Korea. His current research interests include compression algorithm, light-weight networks and h/w accelerators for deep learning models.

Byeongwook Kim (Samsung Research)
Yongkweon Jeon (Samsung Research)
Baeseong Park (samsung research)
Jeongin Yun (Samsung Research)