Search

Accurate recognition of rock fractures is an important problem in rock engineering because fractures greatly influence the mechanical and hydraulic properties of rock structures. However, existing image segmentation methods for identifying rock fractures tend to be limited to handling only very simple fracture images, despite many real cases containing interfering objects or features such as dark surfaces, stripes (e.g., from foliation), infilling materials, scratches, shadows, and vegetation. Here, we propose a novel deep convolutional neural network to construct the first model that is applicable in the field. After selecting U-Net, a simple and powerful network for semantic segmentation, as a baseline network, we tested network architectures by applying atrous convolutions and extra skip connections to develop an optimal network specialized for rock fracture segmentation. The rate of erroneously detecting non-fracture objects or features was reduced by using the atrous convolution module, and more skip connections were appropriately added to increase the detection rate of the actual fractures. The model's performance gradually improved as these new techniques were added to the original model. Contrast-limited adaptive histogram equalization and a fully connected conditional random field were applied before and after the network, respectively, to enhance the model’s performance. Evaluation of the proposed model using raw images of diverse site conditions shows that it can effectively distinguish rock fractures from various interfering objects and features. The source code and pre-trained model can be freely download from GitHub repository ( https://github.com/Montherapy/Rock-fracture-segmentation-with-Tensorflow ).

This paper proposes a deep learning-based denoising method for noisy low-dose computerized tomography (CT) images in the absence of paired training data. The proposed method uses a fidelity-embedded generative adversarial network (GAN) to learn a denoising function from unpaired training data of low-dose CT (LDCT) and standard-dose CT (SDCT) images, where the denoising function is the optimal generator in the GAN framework. This paper analyzes the f-GAN objective to derive a suitable generator that is optimized by minimizing a weighted sum of two losses: the Kullback-Leibler divergence between an SDCT data distribution and a generated distribution, and the $\ell _{2}$ loss between the LDCT image and the corresponding generated images (or denoised image). The computed generator reflects the prior belief about SDCT data distribution through training. We observed that the proposed method allows the preservation of fine anomalous features while eliminating noise. The experimental results show that the proposed deep-learning method with unpaired datasets performs comparably to a method using paired datasets. A clinical experiment was also performed to show the validity of the proposed method for noise arising in the low-dose X-ray CT.

Finding an optimal learning trajectory is an important question in educational systems. Existing Artificial Intelligence in Education (AiEd) technologies mostly used indirect methods to make the learning process efficient such as recommending contents based on difficulty adjustment, weakness analysis, learning theory, psychometric analysis, or domain specific rules. In this study, we propose a recommender system that optimizes the learning trajectory of a student preparing for a standardized exam by recommending the learning content(question) which directly maximizes the expected score after the consumption of the content. In particular, the proposed RCES model computes the expected score of a user by effectively capturing educational effects. To validate the proposed model in an end-to-end system, we conduct an A/B test on 1713 real students by deploying 4 recommenders to a real mobile application. Result shows that RCES has better educational efficiencies than traditional methods such as expert designed models and item response theory based models.

The inevitable shift towards online learning due to the emergence of the COVID-19 pandemic triggered a strong need to assess students using shorter exams whilst ensuring reliability. This study explores a data-centric approach that utilizes feature importance to select a discriminative subset of questions from the original exam. Furthermore, the discriminative question subset’s ability to approximate the students exam scores is evaluated by measuring the prediction accuracy and by quantifying the error interval of the prediction. The approach was evaluated using two real-world exam datasets of the Scholastic Aptitude Test (SAT) and Exame Nacional do Ensino Medio (ENEM) exams, which consist of student response data and the corresponding the exam scores. The evaluation was conducted against randomized question subsets of sizes 10, 20, 30 and 50. The results show that our method estimates the full scores more accurately than a baseline model in most question sizes while maintaining a reasonable error interval. The encouraging evidence found in this paper provides support for the strong potential of the on-going study to provide a data-centric approach for exam size reduction.

Knowledge tracing, the act of modeling a student's knowledge through learning activities, is an extensively studied problem in the field of computer-aided education. Although models with attention mechanism have outperformed traditional approaches such as Bayesian knowledge tracing and collaborative filtering, they share two limitations. Firstly, the models rely on shallow attention layers and fail to capture complex relations among exercises and responses over time. Secondly, different combinations of queries, keys and values for the self-attention layer for knowledge tracing were not extensively explored. Usual practice of using exercises and interactions (exercise-response pairs) as queries and keys/values respectively lacks empirical support. In this paper, we propose a novel Transformer based model for knowledge tracing, SAINT: Separated Self-AttentIve Neural Knowledge Tracing. SAINT has an encoder-decoder structure where exercise and response embedding sequence separately enter the encoder and the decoder respectively, which allows to stack attention layers multiple times. To the best of our knowledge, this is the first work to suggest an encoder-decoder model for knowledge tracing that applies deep self-attentive layers to exercises and responses separately. The empirical evaluations on a large-scale knowledge tracing dataset show that SAINT achieves the state-of-the-art performance in knowledge tracing with the improvement of AUC by 1.8% compared to the current state-of-the-art models., Comment: L@S 2020

A new semi-automated method is proposed in this study that detects joint traces from digital images and calculates the length distribution through three-dimensional data structuring. The method comprises detecting the pixels corresponding to the joint trace in digital images and calculating the length distribution through three-dimensional data structuring of the pixels detected as the joint trace. Semantic segmentation based on deep learning techniques is applied to the detection of joint trace pixels to overcome the limitations of rule-based image processing techniques. For training a deep learning network, various rock joint images were sampled and labeled, following which the data sets for training, validation, and testing were prepared through classification and augmentation. Using the prepared datasets, the performance of the classifiers based on widely used semantic segmentation networks, namely U-Net (University of Freiburg, Germany) network and DeepLabV3+ (Google, USA) network, was compared. The classifier based on the DeepLabV3+ network was ultimately selected and applied. The trained classifier successfully detected pixels corresponding to the joint trace in images of both flat and uneven rock surfaces, indicating that the deep learning technique could be applied effectively to joint trace detection from digital images. Further, a data structuring technique is proposed to calculate the joint trace length using pixelwise data detected through the trained classifier. This technique uses point cloud data obtained by photogrammetry, and comprises two-dimensional thinning and segmentation, three-dimensional projection, segmentation, and segment linking. The entire proposed method, from detecting the joint trace pixels to three-dimensional data structuring, was verified by applying it to both simple flat and uneven rock surface models.

With advances in Artificial Intelligence in Education (AIEd) and the ever-growing scale of Interactive Educational Systems (IESs), data-driven approach has become a common recipe for various tasks such as knowledge tracing and learning path recommendation. Unfortunately, collecting real students' interaction data is often challenging, which results in the lack of public large-scale benchmark dataset reflecting a wide variety of student behaviors in modern IESs. Although several datasets, such as ASSISTments, Junyi Academy, Synthetic and STATICS, are publicly available and widely used, they are not large enough to leverage the full potential of state-of-the-art data-driven models and limits the recorded behaviors to question-solving activities. To this end, we introduce EdNet, a large-scale hierarchical dataset of diverse student activities collected by Santa, a multi-platform self-study solution equipped with artificial intelligence tutoring system. EdNet contains 131,441,538 interactions from 784,309 students collected over more than 2 years, which is the largest among the ITS datasets released to the public so far. Unlike existing datasets, EdNet provides a wide variety of student actions ranging from question-solving to lecture consumption and item purchasing. Also, EdNet has a hierarchical structure where the student actions are divided into 4 different levels of abstractions. The features of EdNet are domain-agnostic, allowing EdNet to be extended to different domains easily. The dataset is publicly released under Creative Commons Attribution-NonCommercial 4.0 International license for research purposes. We plan to host challenges in multiple AIEd tasks with EdNet to provide a common ground for the fair comparison between different state of the art models and encourage the development of practical and effective methods., AIED 2020