Weakly Supervised Concept Map Generation through Task-Guided Graph Translation.

Recent years have witnessed the rapid development of concept map generation techniques due to their advantages in providing well-structured summarization of knowledge from free texts. Traditional unsupervised methods do not generate task-oriented concept maps, whereas deep generative models require large amounts of training data. In this work, we present GT-D2G (Graph Translation-based Document To Graph), an automatic concept map generation framework that leverages generalized NLP pipelines to derive semantic-rich initial graphs, and translates them into more concise structures under the weak supervision of downstream task labels. The concept maps generated by GT-D2G can provide interpretable summarization of structured knowledge for the input texts, which are demonstrated through human evaluation and case studies on three real-world corpora. Further experiments on the downstream task of document classification show that GT-D2G beats other concept map generation methods. Moreover, we specifically validate the labeling efficiency of GT-D2G in the label-efficient learning setting and the flexibility of generated graph sizes in controlled hyper-parameter studies.

Cross-vender, cross-tracer, and cross-protocol deep transfer learning for attenuation map generation of cardiac SPECT.

It has been proved feasible to generate attenuation maps (µ-maps) from cardiac SPECT using deep learning. However, this assumed that the training and testing datasets were acquired using the same scanner, tracer, and protocol. We investigated a robust generation of CT-derived µ-maps from cardiac SPECT acquired by different scanners, tracers, and protocols from the training data. We first pre-trained a network using 120 studies injected with 99mTc-tetrofosmin acquired from a GE 850 SPECT/CT with 360-degree gantry rotation, which was then fine-tuned and tested using 80 studies injected with 99mTc-sestamibi acquired from a Philips BrightView SPECT/CT with 180-degree gantry rotation. The error between ground-truth and predicted µ-maps by transfer learning was 5.13 ± 7.02%, as compared to 8.24 ± 5.01% by direct transition without fine-tuning and 6.45 ± 5.75% by limited-sample training. The error between ground-truth and reconstructed images with predicted µ-maps by transfer learning was 1.11 ± 1.57%, as compared to 1.72 ± 1.63% by direct transition and 1.68 ± 1.21% by limited-sample training. It is feasible to apply a network pre-trained by a large amount of data from one scanner to data acquired by another scanner using different tracers and protocols, with proper transfer learning.

Bayesian models of human navigation behaviour in an augmented reality audiomaze.

We investigated Bayesian modelling of human whole-body motion capture data recorded during an exploratory real-space navigation task in an "Audiomaze" environment (see the companion paper by Miyakoshi et al. in the same volume) to study the effect of map learning on navigation behaviour. There were three models, a feedback-only model (no map learning), a map resetting model (single-trial limited map learning), and a map updating model (map learning accumulated across three trials). The estimated behavioural variables included step sizes and turning angles. Results showed that the estimated step sizes were constantly more accurate using the map learning models than the feedback-only model. The same effect was confirmed for turning angle estimates, but only for data from the third trial. We interpreted these results as Bayesian evidence of human map learning on navigation behaviour. Furthermore, separating the participants into groups of egocentric and allocentric navigators revealed an advantage for the map updating model in estimating step sizes, but only for the allocentric navigators. This interaction indicated that the allocentric navigators may take more advantage of map learning than do egocentric navigators. We discuss relationships of these results to simultaneous localization and mapping (SLAM) problem.

Access-Point Centered Window-Based Radio-Map Generation Network.

Fingerprinting is the term used to describe a common indoor radio-mapping positioning technology that tracks moving objects in real time. To use this, a substantial number of measurement processes and workflows are needed to generate a radio-map. Accordingly, to minimize costs and increase the usability of such radio-maps, this study proposes an access-point (AP)-centered window (APCW) radio-map generation network (RGN). The proposed technique extracts parts of a radio-map in the form of a window based on AP floor plan coordinates to shorten the training time while enhancing radio-map prediction accuracy. To provide robustness against changes in the location of the APs and to enhance the utilization of similar structures, the proposed RGN, which employs an adversarial learning method and uses the APCW as input, learns the indoor space in partitions and combines the radio-maps of each AP to generate a complete map. By comparing four learning models that use different data structures as input based on an actual building, the proposed radio-map learning model (i.e., APCW-based RGN) obtains the highest accuracy among all models tested, yielding a root-mean-square error value of 4.01 dBm.

MapGAN: An Intelligent Generation Model for Network Tile Maps.

In recent years, the generative adversarial network (GAN)-based image translation model has achieved great success in image synthesis, image inpainting, image super-resolution, and other tasks. However, the images generated by these models often have problems such as insufficient details and low quality. Especially for the task of map generation, the generated electronic map cannot achieve effects comparable to industrial production in terms of accuracy and aesthetics. This paper proposes a model called Map Generative Adversarial Networks (MapGAN) for generating multitype electronic maps accurately and quickly based on both remote sensing images and render matrices. MapGAN improves the generator architecture of Pix2pixHD and adds a classifier to enhance the model, enabling it to learn the characteristics and style differences of different types of maps. Using the datasets of Google Maps, Baidu maps, and Map World maps, we compare MapGAN with some recent image translation models in the fields of one-to-one map generation and one-to-many domain map generation. The results show that the quality of the electronic maps generated by MapGAN is optimal in terms of both intuitive vision and classic evaluation indicators.

Smart Wireless Acoustic Sensor Network Design for Noise Monitoring in Smart Cities.

This Special Issue is focused on all the technologies necessary for the development of an efficient wireless acoustic sensor network, from the first stages of its design to the tests conducted during deployment; its final performance; and possible subsequent implications for authorities in terms of the definition of policies. This Special Issue collects the contributions of several LIFE and H2020 projects aimed at the design and implementation of intelligent acoustic sensor networks, with a focus on the publication of good practices for the design and deployment of intelligent networks in any locations.

Automatic Clustering and Classification of Coffee Leaf Diseases Based on an Extended Kernel Density Estimation Approach.

The current methods of classifying plant disease images are mainly affected by the training phase and the characteristics of the target dataset. Collecting plant samples during different leaf life cycle infection stages is time-consuming. However, these samples may have multiple symptoms that share the same features but with different densities. The manual labelling of such samples demands exhaustive labour work that may contain errors and corrupt the training phase. Furthermore, the labelling and the annotation consider the dominant disease and neglect the minor disease, leading to misclassification. This paper proposes a fully automated leaf disease diagnosis framework that extracts the region of interest based on a modified colour process, according to which syndrome is self-clustered using an extended Gaussian kernel density estimation and the probability of the nearest shared neighbourhood. Each group of symptoms is presented to the classifier independently. The objective is to cluster symptoms using a nonparametric method, decrease the classification error, and reduce the need for a large-scale dataset to train the classifier. To evaluate the efficiency of the proposed framework, coffee leaf datasets were selected to assess the framework performance due to a wide variety of feature demonstrations at different levels of infections. Several kernels with their appropriate bandwidth selector were compared. The best probabilities were achieved by the proposed extended Gaussian kernel, which connects the neighbouring lesions in one symptom cluster, where there is no need for any influencing set that guides toward the correct cluster. Clusters are presented with an equal priority to a ResNet50 classifier, so misclassification is reduced with an accuracy of up to 98%.

Multivendor fully automatic uncertainty management approaches for the intuitive representation of DME fluid accumulations in OCT images.

Diabetes represents one of the main causes of blindness in developed countries, caused by fluid accumulations in the retinal layers. The clinical literature defines the different types of diabetic macular edema (DME) as cystoid macular edema (CME), diffuse retinal thickening (DRT), and serous retinal detachment (SRD), each with its own clinical relevance. These fluid accumulations do not present defined borders that facilitate segmentational approaches (specially the DRT type, usually not taken into account by the state of the art for this reason) so a diffuse paradigm is used for its detection and visualization. In this paper, we propose three novel approaches for the representation and characterization of these types of DME. A baseline proposal, using a convolutional neural network as backbone, another based on transfer learning from a general domain, and a third approach exploiting information of regions without a defined label. Overall, our baseline proposal obtained an AUC of 0.9583 ± 0.0093, the approach pretrained with a general-domain dataset an AUC of 0.9603 ± 0.0087, and the approach pretrained in the domain taking advantage of uncertainty, an AUC of 0.9619 ± 0.0073.