🤖 AI Summary
To address the challenge of deploying deep learning models on embedded systems lacking floating-point hardware, this paper proposes the first end-to-end integer-only training and inference framework for arbitrarily deep CNNs—requiring no external quantization or floating-point assistance. Methodologically, it introduces (1) an Integer Local Error Signal (LES) propagation mechanism; (2) the NITRO Scaling learnable scaling layer and NITRO-ReLU activation function; and (3) IntegerSGD, a local-error-driven integer optimizer. All components—including quantized training, gradient updates, and nonlinear activations—are fully integer-native, eliminating floating-point operations entirely. Experiments demonstrate that integer-only MLPs achieve 2.47–5.96% higher accuracy than prior integer baselines, while integer-only CNNs incur only 0.15–4.22% accuracy degradation relative to their floating-point LES counterparts—closely approaching full-precision performance. This work establishes a viable pathway for efficient, hardware-friendly deep learning deployment in resource-constrained environments.
📝 Abstract
Quantization has become increasingly pivotal in addressing the steadily increasing computational and memory requirements of Deep Neural Networks (DNNs). By reducing the number of bits used to represent weights and activations (typically from 32-bit floating-point to 16-bit or 8-bit integers), quantization reduces the memory footprint, energy consumption, and execution time of DNN models. However, traditional quantization methods typically focus on the inference of DNNs, while the training process still relies on floating-point operations. To date, only one work in the literature has addressed integer-only training for Multi-Layer Perceptron (MLP) architectures. This work introduces NITRO-D, a new framework for training arbitrarily deep integer-only Convolutional Neural Networks (CNNs) that operate entirely in the integer-only domain for both training and inference. NITRO-D is the first framework in the literature enabling the training of integer-only CNNs without the need to introduce a quantization scheme. Specifically, NITRO-D introduces a novel architecture integrating multiple integer local-loss blocks, which include the proposed NITRO Scaling Layer and the NITRO-ReLU activation function. Additionally, it introduces a novel integer-only learning algorithm derived from Local Error Signals (LES), utilizing IntegerSGD, an optimizer specifically designed to operate in an integer-only context. NITRO-D is implemented in an open-source Python library. Extensive experimental evaluations demonstrate its effectiveness across several state-of-the-art image recognition datasets. Results show significant performance improvements from 2.47% to 5.96% for integer-only MLP architectures over the state-of-the-art solution, and the capability of training integer-only CNN architectures with minimal accuracy degradation from -0.15% to -4.22% compared to floating-point LES.