WrapNet: Neural Net Inference with Ultra-Low-Resolution Arithmetic
arXiv:2007.13242
Abstract
Low-resolution neural networks represent both weights and activations with few bits, drastically reducing the multiplication complexity. Nonetheless, these products are accumulated using high-resolution (typically 32-bit) additions, an operation that dominates the arithmetic complexity of inference when using extreme quantization (e.g., binary weights). To further optimize inference, we propose a method that adapts neural networks to use low-resolution (8-bit) additions in the accumulators, achieving classification accuracy comparable to their 32-bit counterparts. We achieve resilience to low-resolution accumulation by inserting a cyclic activation layer, as well as an overflow penalty regularizer. We demonstrate the efficacy of our approach on both software and hardware platforms.
References in corpus (8)
- Trained Ternary Quantization
- Training Deep Neural Networks with 8-bit Floating Point Numbers
- WRPN: Wide Reduced-Precision Networks
- Apprentice: Using Knowledge Distillation Techniques To Improve Low-Precision Network Accuracy
- ImageNet pre-trained models with batch normalization
- XNOR-Net++: Improved Binary Neural Networks
- Learning Accurate Low-Bit Deep Neural Networks with Stochastic Quantization
- Quantization of Deep Neural Networks for Accumulator-constrained Processors