LSQ+: Improving low-bit quantization through learnable offsets and better initialization

Abstract

Unlike ReLU, newer activation functions (like Swish, H-swish, Mish) that are frequently employed in popular efficient architectures can also result in negative activation values, with skewed positive and negative ranges. Typical learnable quantization schemes [PACT, LSQ] assume unsigned quantization for activations and quantize all negative activations to zero which leads to significant loss in performance. Naively using signed quantization to accommodate these negative values requires an extra sign bit which is expensive for low-bit (2-, 3-, 4-bit) quantization. To solve this problem, we propose LSQ+, a natural extension of LSQ, wherein we introduce a general asymmetric quantization scheme with trainable scale and offset parameters that can learn to accommodate the negative activations. Gradient-based learnable quantization schemes also commonly suffer from high instability or variance in the final training performance, hence requiring a great deal of hyper-parameter tuning to reach a satisfactory performance. LSQ+ alleviates this problem by using an MSE-based initialization scheme for the quantization parameters. We show that this initialization leads to significantly lower variance in final performance across multiple training runs. Overall, LSQ+ shows state-of-the-art results for EfficientNet and MixNet and also significantly outperforms LSQ for low-bit quantization of neural nets with Swish activations (e.g.: 1.8% gain with W4A4 quantization and upto 5.6% gain with W2A2 quantization of EfficientNet-B0 on ImageNet dataset). To the best of our knowledge, ours is the first work to quantize such architectures to extremely low bit-widths.

Camera-ready for Joint Workshop on Efficient Deep Learning in Computer Vision, CVPR 2020

References in corpus (16)

• EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• Searching for Activation Functions
• Quantizing deep convolutional networks for efficient inference: A whitepaper
• PACT: Parameterized Clipping Activation for Quantized Neural Networks
• Convolutional Neural Networks using Logarithmic Data Representation
• MixConv: Mixed Depthwise Convolutional Kernels
• Learned Step Size Quantization
• Model compression via distillation and quantization
• Apprentice: Using Knowledge Distillation Techniques To Improve Low-Precision Network Accuracy
• Resiliency of Deep Neural Networks under Quantization
• Soft Weight-Sharing for Neural Network Compression
• Trained Quantization Thresholds for Accurate and Efficient Fixed-Point Inference of Deep Neural Networks
• Mixed Precision DNNs: All you need is a good parametrization
• QKD: Quantization-aware Knowledge Distillation
• ZeroQ: A Novel Zero Shot Quantization Framework