Filters

2020 - 20243
Reset filters

Settings

Has files: true × Subject: Deep Learning ×

Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    ItemOpen Access
    Real-time Sound Anomaly Detection in Industrial Environments with Deep Learning
    (Faculty of Engineering and Natural Science, 2024) Zhailau M.
    This research uses deep learning to explore the field of sound anomaly detection in industrial settings in response to the growing need for improved industrial efficiency and safety. Centered on taking care of the constraints of conventional techniques, the study examines the effectiveness of several deep learning architectures, such as hybrid models, recurrent neural networks (RNNs), and convolutional neural networks (CNNs), in identifying abnormal noises. With a focus on rigorous evaluation of datasets, preprocessing methods, and benchmarks, the survey offers a thorough picture of the most recent models and their uses in a variety of industrial areas. This research compares deep learning with traditional methods for sound anomaly identification and looks at performance evaluation criteria. Case studies and realworld implementations demonstrate the usefulness of the enhancements. While highlighting the need for innovative approaches to enhance the practical usefulness and robustness of deep learning-based sound anomaly detection in industrial settings, the research also points out its shortcomings and makes recommendations for future directions. This research not only contributes valuable insights into the intersection of deep learning and industrial sound analysis but also serves as a pivotal guide for researchers and practitioners seeking to navigate the complexities of deploying effective sound anomaly detection systems.
  • Thumbnail Image
    ItemOpen Access
    KAZAKH NAMES GENERATOR USING DEEP LEARNING
    (ВЕСТНИК КАЗАХСТАНСКО-БРИТАНСКОГО ТЕХНИЧЕСКОГО УНИВЕРСИТЕТА, №4, 2020) Nurmambetov D.; Dauylov S.; Bogdanchikov A.
    In recent years, sentiment analysis of e-mail messages or social media posts is becoming very popular. It can help people define if they are reading something positive or negative. On the same time, there are some services on the Internet that can help you find or create a new name. When processing the creation, they check the name in other popular languages, so your name does not mean inappropriate things in other languages. For this they bill for 25 thousand US dollars. If there are such services, then there is a demand. In this study, sentiment analysis of e-mails was implemented with using StanfordNLP [1] lemmatizer and classic machine learning algorithms as a classifier. It is applied to real e-mails from Russian speaking mailbox, which means there are both English and Russian messages. Thus, language identification is also added as preprocessing step. In this study only binary sentiment analysis was made, but it can be improved with adding several emotions to be detected. Then another model generates Kazakh names using neural networks, where all Kazakh names data has been collected through various websites. The sentiment analysis model gives 81% accuracy and the joint use of two models allow us to generate new Kazakh names, which are checked with Russian language if they mean something inappropriate. The result can be improved with checking with other languages.
  • Thumbnail Image
    ItemOpen Access
    FACE-RECOGNITION TO AUTHENTICATE STUDENTS
    (СДУ хабаршысы - 2021, 2021) Yerlan M.
    Abstract. Within the framework of this project, a face recognition system is being developed, which will be used in educational institutions to identify students who are taking exams. To achieve both high quality and fast results, I focused on deep learning approaches to face and object detection and recognition. This research is mainly aimed at providing neural networks and other models with enough data to achieve the desired results. Starting with the basics of neural networks, in which I described and explored a neuron, the smallest unit of deep learning, I brought my research to the point where I could detect a person’s face or an object in a photograph. This research began with the development of neural networks and went on to train them on both the CPU and GPU. In a technique called matrix backpropagation, multiple GPUs were used in conjunction with the CUDA core and the cuBLAS library. Face identification was performed using a pretrained Facenet model combined with deep convolutional neural networks. Numerous ap- proaches to deep learning have been developed from the study of neural networks and their application to face recognition.