Fast neural networks without multipliers

By M. Marchesi, G. Orlandi, Francesco Piazza, Aurelio Uncini. 1993

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Zheng Xu
2. Yunhui Guo
3. Soham De
4. Tom Goldstein
5. Yoshua Bengio
6. Marcel Keller
7. Hao Li
8. Dacheng Tao

Citations in the corpus

• Secure Evaluation of Quantized Neural Networks (2019, background): “4 This is a long-standing research area, with initial works already dating back to the 1990s [5, 26, 49, 62], and this extensive research body have enabled modern quantized neural networks to have essentially the same accuracy as their full-precision counterparts [14, 29, 32, 53, 65].”
• Training Quantized Nets: A Deeper Understanding (2017, background): “Quantized weights can dramatically reduce memory size and access bandwidth, increase power efficiency [1], exploit hardware-friendly bitwise operations [2], and accelerate inference throughput [3, 4].”
• Fully Quantized Transformer for Machine Translation (2019, background): “Consequently, numerical values can be represented using fewer bits (Tang and Kwan, 1993; Marchesi et al., 1993).”
• Backpropagation without Multiplication (1993, methodology): “hese weights in the backward pass, while the weights with reduced resolution are used in the forward pass (Marchesi et al., 1993) .”
• Structure-Adaptable Digital Neural Networks (1999, background): “Another typical approach has been the use of weight and learning parameters values limited to powers of two (thereby reducing multiplications to simple shifts) [46, 132].”
• Extremely Low Bit Neural Network: Squeeze the Last Bit Out with ADMM (2017, background): “The research of low bits quantization of neural network can be traced back to 1990s (Fiesler, Choudry, and Caulfield 1990; Marchesi et al. 1993).”
• A Survey on Methods and Theories of Quantized Neural Networks (2018, background): “Quantizing neural networks dates back to the 1990s [ Fiesler et al. , 1990; Balzer et al. , 1991; Tang and Kwan, 1993; Marchesi et al. , 1993 ] .”
• LightNN: Filling the Gap between Conventional Deep Neural Networks and Binarized Networks (2017, background): “proposed the idea of replacing multipliers of multi-layer perceptrons (MLP) with shifts [16].”