A Survey of Deep Active Learning

Active learning (AL) attempts to maximize a model’s performance gain while annotating the fewest samples possible. Deep learning (DL) is greedy for data and requires a large amount of data supply to optimize a massive number of parameters if the model is to learn how to extract high-quality features. In recent years, due to the rapid development of internet technology, we have entered an era of information abundance characterized by massive amounts of available data. As a result, DL has attracted significant attention from researchers and has been rapidly developed. Compared with DL, however, researchers have a relatively low interest in AL. This is mainly because before the rise of DL, traditional machine learning requires relatively few labeled samples, meaning that early AL is rarely according the value it deserves. Although DL has made breakthroughs in various fields, most of this success is due to a large number of publicly available annotated datasets. However, the acquisition of a large number of high-quality annotated datasets consumes a lot of manpower, making it unfeasible in fields that require high levels of expertise (such as speech recognition, information extraction, medical images, etc.). Therefore, AL is gradually coming to receive the attention it is due. It is therefore natural to investigate whether AL can be used to reduce the cost of sample annotation while retaining the powerful learning capabilities of DL. As a result of such investigations, deep active learning (DeepAL) has emerged. Although research on this topic is quite abundant, there has not yet been a comprehensive survey of DeepAL-related works; accordingly, this article aims to fill this gap. We provide a formal classification method for the existing work, along with a comprehensive and systematic overview. In addition, we also analyze and summarize the development of DeepAL from an application perspective. Finally, we discuss the confusion and problems associated with DeepAL and provide some possible development directions.

Efficient Breast Cancer Classification Using Histopathological Images and a Simple VGG

Breast cancer is the second most deadly disease worldwide. This severe condition led to 627,000 people dying in 2018. Thus, early detection is critical for improving the patients' lifetime or even curing them. In this context, we can appeal to Medicine 4.0, which exploits machine learning capabilities to obtain a faster and more efficient diagnosis. Therefore, this work aims to apply a simpler convolutional neural network, called VGG-7, for classifying breast cancer in histopathological images. Results have shown that VGG-7 overcomes the performance of VGG-16 and VGG-19, showing an accuracy of 98%, a precision of 99%, a recall of 98%, and an F1 score of 98%.

Brain-inspired global-local learning incorporated with neuromorphic computing

There are two principle approaches for learning in artificial intelligence: error-driven global learning and neuroscience-oriented local learning. Integrating them into one network may provide complementary learning capabilities for versatile learning scenarios. At the same time, neuromorphic computing holds great promise, but still needs plenty of useful algorithms and algorithm-hardware co-designs to fully exploit its advantages. Here, we present a neuromorphic global-local synergic learning model by introducing a brain-inspired meta-learning paradigm and a differentiable spiking model incorporating neuronal dynamics and synaptic plasticity. It can meta-learn local plasticity and receive top-down supervision information for multiscale learning. We demonstrate the advantages of this model in multiple different tasks, including few-shot learning, continual learning, and fault-tolerance learning in neuromorphic vision sensors. It achieves significantly higher performance than single-learning methods. We further implement the model in the Tianjic neuromorphic platform by exploiting algorithm-hardware co-designs and prove that the model can fully utilize neuromorphic many-core architecture to develop hybrid computation paradigm.

Monthly Average Rainfall Forecasting Based On An Adaptive Neuro-Fuzzy Inference System in Upper Brahmani Basin, Odisha, India

Floods disrupt human activities, resulting in the loss of lives and property of a region. Excessive rainfall is one of the reasons for flooding, especially in the downstream areas of a catchment. Because of its complexity, understanding and forecasting rainfall is incredibly a challenge. This study investigates the use of an Adaptive Neuro-Fuzzy Inference System (ANFIS) in predicting rainfall using several surface weather parameters as predictors. An ANFIS model is developed for forecasting rainfall over the Upper Brahmani Basin by using 30 years of climate data. A hybrid model with six membership functions gives the best forecast for an area. The suggested method blends neural network learning capabilities with language representations of fuzzy systems that are transparent. The application of ANFIS is to the upper Brahmani river basin is tried for the first time. The ANFIS model with various input structures and membership functions has been built, trained, and tested to evaluate the capability of the model. Statistical performance indices are used to evaluate the performance. Using the developed model, forecast is done for year 2021 – 2030.

Oral language of school-aged children born pretermaturely: a population-based analysis from Madeira Island, Portugal

Premature birth and low birth weight are very important factors in neurodevelopment. Current research in this population focuses on children born prematurely, with no underlying complications in the post-natal period, who are likely to develop specific disorders with their language development and consequently with their learning capabilities too. This study aims to analyse the oral language skills of prematurely born children in comparison to their school-aged peers. The children were assessed in the respective schools, 27 preterm children (16 under 32 weeks and 11 with 32 or more weeks of gestation) and 49 term paired by gender, age, and school year. Tests including simple and complex structures for assessing semantics, morphosyntax, and phonology were used, as well as a test of verbal memory. Preterm born children, regardless of their prematurity grade, showed significantly lower results than their peers, and more than a half of them, 52%, presented low scores in all language tests simultaneously, showing an important language deficit. In contrast, in the term born children group only 14% showed low scores simultaneously in all tests. Verbal memory ability proved to be lower than that of their term peers, regardless of the gestational age and birth weight of preterm children. As a result of this analysis we consider that the evaluation of the linguistic development of these children, even in cases of moderate to late prematurity, should be monitored in order to identify earlier the existence of deficits and prevent psychosocial and learning problems.

Applying a Developmental Lens to Educational Game Designs for Preschoolers

Preschool-aged learners process information differently from older individuals, making it critical to design digital educational games that are tailored to capitalize on young children's learning capabilities. This in-depth literature synthesis connects features of digital educational game design - including visuals, feedback, scaffolding challenge, rewards, and physical interactions to how young children learn. Preschoolers' interests and abilities (e.g., limited attention-span, early reading skills, etc.) are different than older users. As such, developmental science should be used to guide the design of educational games from aesthetic decisions that capture preschoolers' initial interest (e.g., meaningful characters) to carefully select end-of-game rewards (e.g., leveling up). This article connects learning and developmental science research to the design of digital educational games, offering insights into how best to design games for young users and how to select developmentally appropriate games for children.

Implementation of the 2012 upper secondary school curriculum in Mexico: a 21st-century framework enquiry

Objective. This study aims to analyze how the new upper secondary school curriculum in Mexico captures 21st-century skills and teachers’ perceptions of success. Method. The design of the study complies a comparison analysis between the Mexican upper secondary school curriculum and a 21st-century skills framework. Additionally, qualitative data on teachers’ perceptions of success is collected through eight focus groups with 72 participants in 4 States of Mexico. Results. The findings show that the curriculum is short in strategies for the development of 21st-century skills. Moreover, although teachers welcome them, they perceive a lack of support and doubt about students’ learning capabilities. Conclusions. Although Mexico has progressed in providing a 21st-century skills learning environment through the new curriculum, the educational system remains with the opportunity to offer a more suitable and adequate framework as well as support and training for teachers.

On the Neural Network Solution of One-Dimensional Wave Problem

Artificial neural network has become an emerging popular method to handle various problems, especially in case where it has deep multiple neural layers. In this study, we use a deep artificial neural network model to solve one-dimensional wave equation, without any external datasets. Different type of boundary conditions, i.e., Dirichlet, Neumann, and Robin, are used. We analyze the model learning capabilities in a set of settings, such as data setup and the model width and depth. We also present some discussions of advantages and disadvantages of the model in comparison with other matured existing techniques to solve wave equation.

INTERACTIVE MULTIMEDIA ENGLISH LEARNING IN-TEGRATED WITH MOBILE AUGMENTED REALITY

This research mainly focused on multimedia integrated Augmented Reality application that enhances the English learning capabilities of children. Iterative-Cognitive Software development life cycle has been used in this application for its flexibility and changed of adaptability. The application was developed by analysis phase in which the information gathered to identify the problems, scope, objectives and the problem solution. M-Learning offers the great opportunity to learn English language for children at anytime and anywhere by this application. Multimedia integrated M-leaning augmented reality application is the prominent task in this research.

Optimal CNN–Hopfield Network for Pattern Recognition Based on a Genetic Algorithm

Convolutional neural networks (CNNs) have powerful representation learning capabilities by automatically learning and extracting features directly from inputs. In classification applications, CNN models are typically composed of: convolutional layers, pooling layers, and fully connected (FC) layer(s). In a chain-based deep neural network, the FC layers contain most of the parameters of the network, which affects memory occupancy and computational complexity. For many real-world problems, speeding up inference time is an important matter because of the hardware design implications. To deal with this problem, we propose the replacement of the FC layers with a Hopfield neural network (HNN). The proposed architecture combines both a CNN and an HNN: A pretrained CNN model is used for feature extraction, followed by an HNN, which is considered as an associative memory that saves all features created by the CNN. Then, to deal with the limitation of the storage capacity of the HNN, the proposed work uses multiple HNNs. To optimize this step, the knapsack problem formulation is proposed, and a genetic algorithm (GA) is used solve it. According to the results obtained on the Noisy MNIST Dataset, our work outperformed the state-of-the-art algorithms.