The quantization of deep neural networks (QDNNs) has been actively studied for deployment in edge devices. Recent studies employ the knowledge distillation (KD) method to improve the performance of quantized networks. In this study, we propose stochastic precision ensemble training for QDNNs (SPEQ). SPEQ is a knowledge distillation training scheme; however, the teacher is formed by sharing the model parameters of the student network. We obtain the soft labels of the teacher by changing the bit precision of the activation stochastically at each layer of the forward-pass computation. The student model is trained with these soft labels to reduce the activation quantization noise. The cosine similarity loss is employed, instead of the KL-divergence, for KD training. As the teacher model changes continuously by random bit-precision assignment, it exploits the effect of stochastic ensemble KD. SPEQ outperforms the existing quantization training methods in various tasks, such as image classification, question-answering, and transfer learning without the need for cumbersome teacher networks.

Quantized deep neural networks (QDNNs) are necessary for low-power, high throughput, and embedded applications. Previous studies mostly focused on developing optimization methods for the quantization of given models. However, quantization sensitivity depends on the model architecture. Therefore, the model selection needs to be a part of the QDNN design process. Also, the characteristics of weight and activation quantization are quite different. This study proposes a holistic approach for the optimization of QDNNs, which contains QDNN training methods as well as quantization-friendly architecture design. Synthesized data is used to visualize the effects of weight and activation quantization. The results indicate that deeper models are more prone to activation quantization, while wider models improve the resiliency to both weight and activation quantization. This study can provide insight into better optimization of QDNNs.

Knowledge distillation (KD) is a very popular method for model size reduction. Recently, the technique is exploited for quantized deep neural networks (QDNNs) training as a way to restore the performance sacrificed by word-length reduction. KD, however, employs additional hyper-parameters, such as temperature, coefficient, and the size of teacher network for QDNN training. We analyze the effect of these hyper-parameters for QDNN optimization with KD. We find that these hyper-parameters are inter-related, and also introduce a simple and effective technique that reduces \textit{coefficient} during training. With KD employing the proposed hyper-parameters, we achieve the test accuracy of 92.7% and 67.0% on Resnet20 with 2-bit ternary weights for CIFAR-10 and CIFAR-100 data sets, respectively.

Real-time automatic speech recognition (ASR) on mobile and embedded devices has been of great interests for many years. We present real-time speech recognition on smartphones or embedded systems by employing recurrent neural network (RNN) based acoustic models, RNN based language models, and beam-search decoding. The acoustic model is end-to-end trained with connectionist temporal classification (CTC) loss. The RNN implementation on embedded devices can suffer from excessive DRAM accesses because the parameter size of a neural network usually exceeds that of the cache memory and the parameters are used only once for each time step. To remedy this problem, we employ a multi-time step parallelization approach that computes multiple output samples at a time with the parameters fetched from the DRAM. Since the number of DRAM accesses can be reduced in proportion to the number of parallelization steps, we can achieve a high processing speed. However, conventional RNNs, such as long short-term memory (LSTM) or gated recurrent unit (GRU), do not permit multi-time step parallelization. We construct an acoustic model by combining simple recurrent units (SRUs) and depth-wise 1-dimensional convolution layers for multi-time step parallelization. Both the character and word piece models are developed for acoustic modeling, and the corresponding RNN based language models are used for beam search decoding. We achieve a competitive WER for WSJ corpus using the entire model size of around 15MB and achieve real-time speed using only a single core ARM without GPU or special hardware.

Real-time automatic speech recognition (ASR) on mobile and embedded devices has been of great interests for many years. We present real-time speech recognition on smartphones or embedded systems by employing recurrent neural network (RNN) based acoustic models, RNN based language models, and beam-search decoding. The acoustic model is end-to-end trained with connectionist temporal classification (CTC) loss. The RNN implementation on embedded devices can suffer from excessive DRAM accesses because the parameter size of a neural network usually exceeds that of the cache memory and the parameters are used only once for each time step. To remedy this problem, we employ a multi-time step parallelization approach that computes multiple output samples at a time with the parameters fetched from the DRAM. Since the number of DRAM accesses can be reduced in proportion to the number of parallelization steps, we can achieve a high processing speed. However, conventional RNNs, such as long short-term memory (LSTM) or gated recurrent unit (GRU), do not permit multi-time step parallelization. We construct an acoustic model by combining simple recurrent units (SRUs) and depth-wise 1-dimensional convolution layers for multi-time step parallelization. Both the character and word piece models are developed for acoustic modeling, and the corresponding RNN based language models are used for beam search decoding. We achieve a competitive WER for WSJ corpus using the entire model size of around 15MB and achieve real-time speed using only a single core ARM without GPU or special hardware.

We propose a fast approximation method of a softmax function with a very large vocabulary using singular value decomposition (SVD). SVD-softmax targets fast and accurate probability estimation of the topmost probable words during inference of neural network language models. The proposed method transforms the weight matrix used in the calculation of the output vector by using SVD. The approximate probability of each word can be estimated with only a small part of the weight matrix by using a few large singular values and the corresponding elements for most of the words. We applied the technique to language modeling and neural machine translation and present a guideline for good approximation. The algorithm requires only approximately 20\% of arithmetic operations for an 800K vocabulary case and shows more than a three-fold speedup on a GPU.