Automatic processing of emotion information through deep neural networks (DNN) can have great benefits (e.g., for human-machine interaction). Vice versa, machine learning can profit from concepts known from human information processing (e.g., visual attention). We employed a recurrent DNN incorporating a spatial attention mechanism for facial emotion recognition (FER) and compared the output of the network with results from human experiments. The attention mechanism enabled the network to select relevant face regions to achieve state-of-the-art performance on a FER database containing images from realistic settings. A visual search strategy showing some similarities with human saccading behavior emerged when the model’s perceptive capabilities were restricted. However, the model then failed to form a useful scene representation.

京都大学 / 0048 / 新制・課程博士 / 博士(農学) / 甲第22077号 / 農博第2369号 / 新制||農||1072(附属図書館) / 学位論文||R1||N5231(農学部図書室) / 京都大学大学院農学研究科地域環境科学専攻 / (主査)教授 近藤 直, 教授 清水 浩, 教授 飯田 訓久 / 学位規則第4条第1項該当 / Doctor of Agricultural Science / Kyoto University / DGAM

No description available.

<div>Today, machine learning models and Deep Neural Networks (DNNs) are prevalent in various areas. Also, educational Artificial Intelligence (AI) is drawing increasing attention</div><div>with the rapid development of online learning platforms. Researchers explore different types of educational AI to improve students’ learning performance and experience in online classes. Educational AIs can be categorized into “interactive” and “predictive.” Interactive AIs answer simple course questions for students, such as the due day of homework and the final project’s minimum page requirement. Predictive educational AIs play a role in predicting students’ learning states. Instructors can adjust the learning content based on the students’ learning states. However, most AIs are not evaluated in an actual class setting. Therefore, we want to evaluate the effects of a state-of-the-art educational AI model, BKT (Bayesian Knowledge Tracing)-LSTM(Long Short-Term Memory), on students’ learning performance in an actual class setting. Data came from the course CNIT 25501, a large introductory Java program?ming class at Purdue University. Participants were randomly separated into the control and experimental groups (AI-group). Weekly quizzes measured participants’ learning performance. Pre-quiz and base quizzes estimated participants’ prior knowledge levels. Using BKT-LSTM, participants in the experimental group had questions from the knowledge that they were most lacking. However, participants in the control group had questions from randomly picked knowledge. The results suggested that both the experimental and control groups had lower scores in review quizzes than in base quizzes. However, the score difference between base quizzes and review quizzes for the experimental group was more often significantly different (three quizzes) compared to the control group (two quizzes), demonstrating the predictive capability of BKT-LSTM to some extent. Initially, we expected that BKT-LSTM would enhance students’ learning performance. However, in post-quiz, participants in the control group had significantly higher scores than those in the experimental group. The result suggested that continuous complex questions may negatively affect students’ learning initiatives. On the contrary, relatively easy questions may improve their learning initiatives.</div>

<div>With the advent of Machine Learning and Deep Learning models, many avenues of development have opened. Today, these technologies are being leveraged to perform a wide variety of tasks that were otherwise not possible with traditional systems. The power of Machine Learning and Artificial Intelligence makes it possible to compute very complicated tasks at near real-time speeds. To provide an example, Machine Learning models are used extensively in the retail industry to predict and analyze critical parameters such as sales, promotions, customer behavior, recommendations, and offers.</div><div><br></div><div><br></div><div>Today, it is increasingly common to observe AI being used across many of the biggest domains such as Health, Environment, Military, and Business. Artificial Intelligence being used in educational settings has thus been a growing field of focus and study. For example, conversational AI being deployed to act as virtual tutors to answer student questions and concerns. Additionally, there is a fill-the-hole type of AIs that will help students learn tasks such as coding by either showing them how to do it or by predicting where the student might go wrong and suggesting preemptive corrective steps. </div><div><br></div><div>As described, a great deal of literature exists about the use of Deep Learning and Machine Learning models in education. However, the existing tools and models act as external appendages that add to the course structure, thereby altering it. This proposed study introduces a Bayesian Knowledge Transfer model based on the Long Short Term Memory structure (BKT-LSTM) utilized in a live STEM (Science, Technology, Engineering, and Mathematics) classroom. The model discovers individual student learning profiles based on past quiz performance and customizes future quizzes based on the learned patterns. The BKT-LSTM model works in tandem with the existing course curriculum and only tests those knowledge items that have already been covered in the classroom. The model does not change the course structure but rather aims to improve the student’s learning experience by focusing on areas of the student's knowledge that require more practice in learning. </div><div><br></div><div><br></div><div>Within a live STEM classroom, the BKT-LSTM model acts as a herald of change in the way students interact with the curriculum, even though no major changes are observed in the course structure. Students interacting with the model are subjected to quizzes with questions that target the individual student’s lack of learning in particular knowledge areas. Thus, students can be expected to perceive the change as unwelcoming due to the increasing difficulty in subsequent quizzes. Regardless, the study focuses on measuring the learning performance of the students. Do the students learn more in the new system? Another focus of the study is the student’s perception of engagement while interacting with the BKT-LSTM model. The effectiveness of the new educational process is determined not only by increased student learning performance, but also by the student’s perception of engagement while interacting with the model. Are the students enjoying the new experience? Do the students feel like they are learning something? Another important factor was also studied, that is learning performance of students interacting with the BKT-LSTM. </div><div><br></div>

The thesis deals with methods for analysis of financial markets, focused on cryptocurrencies. The theoretical part, in a context of virtual currencies, describes block-chain technology, financial indicators and neural networks with recurrent architectures. Main goal is to create a system for giving a recommendation either for buy, or sell the currency. The system consists of designed financial strategy and predicted value of the currency, for which is used financial indicators and LSTM neural network. Tests were performed on Bitcoin, Litecoin and Ethereum historical data from year 2017.

This thesis is concerned with future trend prediction on capital markets on the basis of neural networks. Usage of convolutional and recurrent neural networks, Elliott wave theory and scalograms for capital market's future trend prediction is discussed. The aim of this thesis is to propose a novel approach to future trend prediction based on Elliott's wave theory. The proposed approach will be based on the principle of classification of chosen patterns from Elliott's theory by the way of convolutional neural network. To this end scalograms of the chosen Elliott patterns will be created through application of continuous wavelet transform on parts of historical time series of price for chosen stocks.

This diploma thesis deals with the detection of fire and smoke from the image signal. The approach of this work uses a combination of convolutional and recurrent neural network. Machine learning models created in this work contain inception modules and blocks of long short-term memory. The research part describes selected models of machine learning used in solving the problem of fire detection in static and dynamic image data. As part of the solution, a data set containing videos and still images used to train the designed neural networks was created. The results of this approach are evaluated in conclusion.

In this master thesis project, four well known time series forecasting models areconstructed and tuned with the purpose of predicting the future consumption of glueon one of AkzoNobels customers production lines. The goal was to examine thepossibility of utilizing their vastly collected data with these models to improve on theinventory management for both AkzoNobel and their customers. The predictedproduct usage rate would aid in the customers' decision making about when neworders of product should be placed, based on when the current storage tanks areforecasted to be emptied. This information could also be useful for AkzoNobelthemselves. The data that is handled in this project is a time series with timestampsfor every glue consumption process on the customers production line since 2017. Asubgoal was to determine what data resolution would be the most effective formodelling, so each model has two versions, one using higher and one using lowerresolution data. The models that are examined are a seasonal naive model,along-short term memory model, a Facebook Prophet model as well as two separateAutoregressive Integrated Moving Average models, specifically one automaticallyandone manually constructed. Beyond these models, a combined model using trueaveraging of the two automatic ARIMA models was examined as well.   Ultimately it was found that, for most models, forecasting ahead with a one day resolution was the most accurate using the models trained on one-day-separated-data, compared to three-hour-separated-data. Further it is presented that the best models are the two naive models, closely followed by the one-day-case automatic ARIMA and Prophet models. These models also performed similarly on simple tests for predicting a date when a tank will be empty. Mostly differing around four days on average from the true date for an empty tank on those tests, with a max forecast range of forty days. It is concluded that it is possible to sufficiently model the data to a point where the best models in this project could be an effective tool for both the AkzoNobel and its customers.

Context: The context of this thesis to predict user mobility using deep learning algorithms which can increase the quality of service for the users and reduce the cost of paging for telecom carriers. Objectives: This study first investigates to find the suitable deep learning algorithms that can be used to predict user mobility and then an experiment is performed with the chosen algorithms as a global model and individual model then evaluate the performance of algorithms. Methods: Firstly, a Literature review is used to find suitable deep learning algorithms and then based on finding an experiment is performed to evaluate the chosen deep learning algorithms. Results: Results from the literature review show that the RNN, LSTM, and variants of the LSTM are the suitable deep learning algorithms. The models are evaluated with metrics accuracy. The results from the experiment showed that the individual model gives better performance in predicting user mobility when compared to the global model. Conclusions: From the results obtained from the experiment, it can be concluded that the individual model is the technique of choice in predicting user mobility.