TENT: Efficient Quantization of Neural Networks on the tiny Edge with Tapered FixEd PoiNT
This work addresses energy-efficient neural network deployment on edge devices, representing an incremental improvement in quantization methods for TinyML.
The researchers tackled the problem of efficiently quantizing neural networks for TinyML by proposing TENT, a low-precision framework using tapered fixed-point quantization, which improved accuracy by up to ~31% with an energy overhead of ~17-30% compared to fixed-point for ConvNet and ResNet-18 models.
In this research, we propose a new low-precision framework, TENT, to leverage the benefits of a tapered fixed-point numerical format in TinyML models. We introduce a tapered fixed-point quantization algorithm that matches the numerical format's dynamic range and distribution to that of the deep neural network model's parameter distribution at each layer. An accelerator architecture for the tapered fixed-point with TENT framework is proposed. Results show that the accuracy on classification tasks improves up to ~31 % with an energy overhead of ~17-30 % as compared to fixed-point, for ConvNet and ResNet-18 models.