Search

Automatic image description, also known as image captioning, aims to describe the elements included in an image and their relationships. This task involves two research fields: computer vision and natural language processing; thus, it has received much attention in computer science. In this review paper, we follow the Kitchenham review methodology to present the most relevant approaches to image description methodologies based on deep learning. We focused on works using convolutional neural networks (CNN) to extract the characteristics of images and recurrent neural networks (RNN) for automatic sentence generation. As a result, 53 research articles using the encoder-decoder approach were selected, focusing only on supervised learning. The main contributions of this systematic review are: (i) to describe the most relevant image description papers implementing an encoder-decoder approach from 2014 to 2022 and (ii) to determine the main architectures, datasets, and metrics that have been applied to image description. [ABSTRACT FROM AUTHOR]

Text-to-speech (TTS) synthesizers have been widely used as a vital assistive tool in various fields. Traditional sequence-to-sequence (seq2seq) TTS such as Tacotron2 uses a single soft attention mechanism for encoder and decoder alignment tasks, which is the biggest shortcoming that incorrectly or repeatedly generates words when dealing with long sentences. It may also generate sentences with run-on and wrong breaks regardless of punctuation marks, which causes the synthesized waveform to lack emotion and sound unnatural. In this paper, we propose an end-to-end neural generative TTS model that is based on the deep-inherited attention (DIA) mechanism along with an adjustable local-sensitive factor (LSF). The inheritance mechanism allows multiple iterations of the DIA by sharing the same training parameter, which tightens the token–frame correlation, as well as fastens the alignment process. In addition, LSF is adopted to enhance the context connection by expanding the DIA concentration region. In addition, a multi-RNN block is used in the decoder for better acoustic feature extraction and generation. Hidden-state information driven from the multi-RNN layers is utilized for attention alignment. The collaborative work of the DIA and multi-RNN layers contributes to outperformance in the high-quality prediction of the phrase breaks of the synthesized speech. We used WaveGlow as a vocoder for real-time, human-like audio synthesis. Human subjective experiments show that the DIA-TTS achieved a mean opinion score (MOS) of 4.48 in terms of naturalness. Ablation studies further prove the superiority of the DIA mechanism for the enhancement of phrase breaks and attention robustness. [ABSTRACT FROM AUTHOR]

At present, short text classification is a hot topic in the area of natural language processing. Due to the sparseness and irregularity of short text, the task of short text classification still faces great challenges. In this paper, we propose a new classification model from the aspects of short text representation, global feature extraction and local feature extraction. We use convolutional networks to extract shallow features from short text vectorization, and introduce a multi-level semantic extraction framework. It uses BiLSTM as the encoding layer while the attention mechanism and normalization are used as the interaction layer. Finally, we concatenate the convolution feature vector and semantic results of the semantic framework. After several rounds of feature integration, the framework improves the quality of the feature representation. Combined with the capsule network, we obtain high-level local information by dynamic routing and then squash them. In addition, we explore the optimal depth of semantic feature extraction for short text based on a multi-level semantic framework. We utilized four benchmark datasets to demonstrate that our model provides comparable results. The experimental results show that the accuracy of SUBJ, TREC, MR and ProcCons are 93.8%, 91.94%, 82.81% and 98.43%, respectively, which verifies that our model has greatly improves classification accuracy and model robustness. [ABSTRACT FROM AUTHOR]

In the last few decades, text mining has been used to extract knowledge from free texts. Applying neural networks and deep learning to natural language processing (NLP) tasks has led to many accomplishments for real-world language problems over the years. The developments of the last five years have resulted in techniques that have allowed for the practical application of transfer learning in NLP. The advances in the field have been substantial, and the milestone of outperforming human baseline performance based on the general language understanding evaluation has been achieved. This paper implements a targeted literature review to outline, describe, explain, and put into context the crucial techniques that helped achieve this milestone. The research presented here is a targeted review of neural language models that present vital steps towards a general language representation model. [ABSTRACT FROM AUTHOR]

Recently, deep learning has achieved state-of-the-art performance in more aspects than traditional shallow architecture-based machine-learning methods. However, in order to achieve higher accuracy, it is usually necessary to extend the network depth or ensemble the results of different neural networks. Increasing network depth or ensembling different networks increases the demand for memory resources and computing resources. This leads to difficulties in deploying depth-learning models in resource-constrained scenarios such as drones, mobile phones, and autonomous driving. Improving network performance without expanding the network scale has become a hot topic for research. In this paper, we propose a cross-architecture online-distillation approach to solve this problem by transmitting supplementary information on different networks. We use the ensemble method to aggregate networks of different structures, thus forming better teachers than traditional distillation methods. In addition, discontinuous distillation with progressively enhanced constraints is used to replace fixed distillation in order to reduce loss of information diversity in the distillation process. Our training method improves the distillation effect and achieves strong network-performance improvement. We used some popular models to validate the results. On the CIFAR100 dataset, AlexNet's accuracy was improved by 5.94%, VGG by 2.88%, ResNet by 5.07%, and DenseNet by 1.28%. Extensive experiments were conducted to demonstrate the effectiveness of the proposed method. On the CIFAR10, CIFAR100, and ImageNet datasets, we observed significant improvements over traditional knowledge distillation. [ABSTRACT FROM AUTHOR]