Search

The actual needs of users for information are often hidden in multiple question answering (QA) on the same topic. In order to generate answers to users' current questions, a conversational QA system relies on getting external information from a knowledge graph (KG) and combining it with conversation history. Therefore, how to make full use of the information of KG and combining it with the conversation environment is the top priority. In a conversational knowledge graph question answering (KGQA) scenario, the follow-up questions are often incomplete and may contain a shift of focal entity. Considering and processing the constrained information which plays the key role in solving complex conversational KGQA is very important. In this paper, we propose a reinforcement learning (RL) model, which uses a dynamically maintained context entity set to capture the shift of the focal entity in the process of conversation. We then use the bidirectional encoder representations from transformers (BERT) pre-training model to obtain the semantic information from context questions and KG paths. Our model learns from not only the 1-hop path but also 2-hop path constraint of the KG at the same time and gives reward rules based on precision and certain rules, respectively. Compared with state-of-the-art methods on ConvQuestions, our model improves mean reciprocal rank (MRR) and precision at 1 (P@1) by a margin of 4.7 % and 4.3 % , respectively. Experimental results on two datasets demonstrate the effectiveness of our proposed approach. [ABSTRACT FROM AUTHOR]

In recent years, deep learning has been widely applied to single image super-resolution(SISR). However, the majority of deep learning methods employ the Mean Square Error(MSE) loss as the objective optimization function, and the generated results are frequently too smooth and lack of details. In addition, the high-frequency information of the reconstructed image is severely lost, resulting in a generated image with poor visual effects. In order to address the aforementioned issues, this paper proposes a super-resolution network (RRFDB-GAN) with a receptive field module with a generative adversarial network as the main framework. The network adopts the receptive field block (RFB), which enables it to extract the features in multiple scales and to improve the discriminability. In this paper, the Residual in Residual Dense Block (RRDB) and the Residual of Receptive Field Dense Block (RRFDB) are combined into a new module, called Basic block. This module enhances the capability of feature reconstruction for low-resolution images. During the upsampling stage, a combination of Nearest Neighborhood Interpolation and Sub-pixel convolution is used to reduce the computational complexity and provide additional contextual information for super-resolution reconstruction, while achieving satisfactory performance. Finally, the four Basic blocks are integrated with the upsampling module into a simple end-to-end framework. Extensive experimental results demonstrate that the proposed method in this paper shows more details on the five test sets and outperforms other methods in terms of quantitative metrics and perception assessment. [ABSTRACT FROM AUTHOR]

Computer vision (CV) technologies are assisting the health care industry in many respects, i.e., disease diagnosis. However, as a pivotal procedure before and after surgery, the inventory work of surgical instruments has not been researched with the CV-powered technologies. To reduce the risk and hazard of surgical tools' loss, we propose a study of systematic surgical instrument classification and introduce a novel attention-based deep neural network called SKA-ResNet which is mainly composed of: (a) A feature extractor with selective kernel attention module to automatically adjust the receptive fields of neurons and enhance the learnt expression and (b) A multi-scale regularizer with KL-divergence as the constraint to exploit the relationships between feature maps. Our method is easily trained end-to-end in only one stage with few additional calculation burdens. Moreover, to facilitate our study, we create a new surgical instrument dataset called SID19 (with 19 kinds of surgical tools consisting of 3800 images) for the first time. Experimental results show the superiority of SKA-ResNet for the classification of surgical tools on SID19 when compared with state-of-the-art models. The classification accuracy of our method reaches up to 97.703%, which is well supportive for the inventory and recognition study of surgical tools. Also, our method can achieve state-of-the-art performance on four challenging fine-grained visual classification datasets. [ABSTRACT FROM AUTHOR]