Hasil Pencarian  ::  Simpan CSV :: Kembali

"ABSTRAKDalam beberapa tahun terakhir, Deep Learning DL telah menarik banyak perhatian dalam penelitian pemelajaran mesin. Metode ini telah berhasil dipakai untuk berbagai aplikasi pada pemrosesan suatu, robotika, pengenalan fonetik, pencarian informasi dan bahkan analisa molekul. Meskipun DL telah berhasil sukses untuk diterapkan dalam berbagai bidang aplikasi, training yang diperlukan pada metode ini tidaklah mudah. Sejumlah cara telah diusulkan untuk membuat proses training DL menjadi lebih optimal, beberapa diantanya dengan menambahkan proses pre-training, memutuskan beberapa jaringan dalam lapisan, ataupun mengganti fungsi aktivasi dan metode gradien standar yang dipergunakan. Disertasi ini menggunakan pendekatan lain dalam optimasi DL, yaitu memakai algoritme metaheuristik. Secara umum disertasi ini dibagi dalam dua bagian besar. Bagian pertama adalah studi awal penelitian yang difokuskan pada beberapa eksperimen yang berkaitan dengan algoritme metaheuristik dan aplikasi DL dalam klasifikasi citra. Bagian kedua dari disertasi berkaitan dengan penerapan algoritme metaheuristik dalam DL. Hasil pada bagian ini misalnya untuk optimasi metode Convolutional Neural Nework CNN menggunakan dataset CIFAR10, diperoleh untuk Top-1 error pada validasi adalah 99,05 . Hasil ini lebih baik dari nilai akurasi CNN asli sebesar 88,21 , fine-tuning CNN menggunakan Harmony Search yang diusulkan G. Rosa dkk sebesar 78,28 , dan bahkan State of the art saat ini sebesar 96,53 dengan Fractional Max-Pooling.

---

ABSTRACTIn recent years, deep Learning DL has drawn many attention in machine learning research. This method has been successfully used in various applications, such as sound process, robotics, phonetic identification, information retrieval, and even molecule analysis. Although DL has been successful to be applied in many fields, it is difficult to train in this method. Various attempts and methods has been proposed to make the DL training process become more optimum, some of them are by adding pre training process, drop out some networks in the layer, or by replacing activation function and standard gradient method being used. This dissertation takes another way to optimize a DL, i.e. using metaheuristic algorithms. Overall, this dissertation will be divided into two main parts. The first part is a preliminary study of research, focusing on several experiments which were related to the metaheuristic algorithm and DL application in image classification. The second part of this dissertation is related to application of metaheuristic algorithm in DL. The results in this part, for example, the optimization of CNN method using CIFAR10 dataset for Top 1 error in validation is 99.05 . This result is higher than the accuracy level from original CNN 88,21 , fine tuning CNN using Harmony Search suggested by G. Rossa et.al 78.28 , and even ldquo State of the art rdquo right now using Fractional Max Pooling 96.53 "

"Pasar modal berkembang pesat di Indonesia dengan peningkatan 79 jumlah emiten saham baru dan peningkatan 17,9% jumlah investor baru. Perkembangan ini dipacu oleh Otoritas Jasa Keuangan yang meyakinkan bahwa setiap perusahaan terbuka selalu diawasi dengan cara mewajibkan perusahaan terbuka untuk menyampaikan laporan keuangan secara berkala. Akan tetapi pada kenyataannya, tindakan kecurangan laporan keuangan bukan menjadi hal yang langka. Association of Certified Fraud Examiner melaporkan bahwa 9,2% kecurangan di Indonesia merupakan kecurangan laporan keuangan dengan total kerugian hingga Rp242.260.000.000. Sementara, proses audit konvensional serta laporan yang menjadi 72% dari media deteksi saat ini membutuhkan 12 bulan untuk mendeteksi kasus kecurangan. Penelitian ini akan menggunakan metode ensemble learning berbasis optimasi metaheuristik untuk mengembangkan model deteksi kecurangan pada laporan keuangan. Beberapa metode klasifikasi digunakan untuk mengembangkan model, yaitu Random Forest dan XGBoost. Optimasi metaheuristik dengan metode Genetic Algorithm kemudian digunakan sebagai dasar dari proses hyperparameter tuning pada model tersebut. Hasil deteksi terbaik pada penelitian ini adalah model XGBoost dengan parameter teroptimasi yang menghasilkan akurasi sebesar 98,04% dan sensitivitas 99.02%.

---

The capital market is growing rapidly in Indonesia, gaining 79 new stock issuers and a 17.9% increase in the number of new investors in 2023. This development is driven by Otoritas Jasa Keuangan, which ensures that every public company is always monitored by requiring them to submit financial statements regularly. However, financial statement fraud is not uncommon. The Association of Certified Fraud Examiners reports that 9.2% of fraud cases in Indonesia involve financial statement fraud, with total losses amounting to Rp242,260,000,000. Meanwhile, conventional audit processes and reports, which account for 72% of current detection methods, take 12 months to detect fraud cases. This study will use an ensemble learning method based on metaheuristic optimization to develop a fraud detection model for financial statements. Several classification methods, namely Random Forest and XGBoost, are used to develop the model. Metaheuristic optimization using the Genetic Algorithm method is then applied as the basis for hyperparameter tuning in this model. The best detection result in this study is achieved by the XGBoost model with optimized parameters, yielding an accuracy of 98.04% and a sensitivity of 99.02%."

"Arsitektur model deep learning kini sudah semakin kompleks setiap harinya. Namun semakin besar model maka dibutuhkan kekuatan komputasi yang cukup besar juga dalam menjalankan model. Sehingga tidak semua perangkat Internet of Things (IoT) dapat menjalankan model yang begitu besar secara efisien. Untuk itu teknik model optimization sangat diperlukan. Pada penelitian kali ini penulis menggunakan metode optimasi menggunakan layer weight regularization, serta penyederhanaan arsitektur model pada hybrid deep learning model. Dataset yang digunakan pada penelitian kali ini adalah N-BaIoT. Sementara evaluasi performa model yang digunakan adalah accuracy, confussion matrix, dan detection time. Dengan tingkat accuracy yang sama, model yang diusulkan berhasil meningkatkan waktu deteksi model lebih cepat 0,8 ms dibandingkan dengan model acuan.

---

The deep learning model architecture is getting more complex every day. However, the larger the model, the greater the computational power is needed to run the model. So not all Internet of Things (IoT) devices can run such a large model efficiently. For this reason, model optimization techniques are needed. In this study, the author uses an optimization method using layer weight regularization, as well as simplification of the model architecture on the hybrid deep learning model. The dataset used in this research is N-BaIoT. While the evaluation of the performance of the model used is accuracy, confusion matrix, and detection time. With the same level of accuracy, the proposed model succeeded in increasing the detection time of the model by 0.8 ms faster than the reference method."

"Seiring perkembangan teknologi, pemanfaatan dari pengembangan ilmu pengetahuan tersebut harus selalu ditingkatkan. Khususnya dalam hal menciptakan keamanan dan ketertiban di Indonesia. Sementara jumlah proporsi polisi dan warga yang tidak ideal, 1:900, Kepolisian Republik Indonesia masih menggunakan cara manual yang tidak efektif dalam mengidentifikasi pelaku kejahatan. Yaitu membuat sketsa wajah pelaku kejahatan dan mencari kemiripan wajah dengan citra-citra wajah yang ada di basis data Kepolisian. Sistem Identifikasi Buron bagian Alis dibuat untuk memperbaiki ketidakefektifan proses tersebut. Sistem Identifikasi Buron bagian Alis merupakan sub-bagian dari sistem Identifikasi Buron yang menggunakan bagian-bagian wajah lainnya untuk proses identifikasi. Untuk mencari yang paling efektif dalam mengukur kemiripan alis, maka penelitian ini membandingkan dua metode yang diganakan untuk melakukan ekstraksi. Yaitu Eigenface dan Klustering K-Means dengan Koreksi Gamma. Selain itu, penelitian ini juga membagi alis menjadi lima kategori, tebal, tipis, sambung, normal, dan sedang. Citra wajah yang digunakan berasal dari citra mahasiswa Universitas Indonesia (UI) angkatan 2007 sebanyak 500 buah. Citra alis diperoleh dari data wajah tersebut yang di crop secara manual. Keseluruhan data ini diperoleh dari Pusat Pengembangan Sistem Informasi (PPSI) UI. Setiap metode akan diuji dengan memberikan lima template dari lima kategori yang berbeda untuk diuji kemiripannya. Dari penelitian ini dihasilkan bahwa Eigenface memiliki akurasi sebesar 64.64%, sedangkan Klustering K-Means dengan Koreksi Gamma memiliki akurasi sebesar 74.75%. Diharapkan hasil penelitian ini bisa membantu kepolisian dalam menjaga keamanan dan ketertiban di Indonesia."

"Learning in non-stationary environments : methods and applications offers a wide-ranging, comprehensive review of recent developments and important methodologies in the field. The coverage focuses on dynamic learning in unsupervised problems, dynamic learning in supervised classification and dynamic learning in supervised regression problems. A later section is dedicated to applications in which dynamic learning methods serve as keystones for achieving models with high accuracy. Rather than rely on a mathematical theorem/proof style, the editors highlight numerous figures, tables, examples and applications, together with their explanations."

"This book includes innovative research work presented at ICO’2018, the 1st International Conference on Intelligent Computing and Optimization, held in Pattaya, Thailand on October 4–5, 2018. The conference presented topics ranging from power quality, reliability, security assurance, cloud computing, smart cities, renewable energy, agro-engineering, smart vehicles, deep learning, block chain, power systems, AI, machine learning, manufacturing systems, and big-data analytics. This volume focuses on subjects related to innovative computing, uncertainty management and optimization approaches to real-world problems in big-data, smart cities, sustainability, meta-heuristics, cyber-security, IoTs, economics and finance, renewable energy, energy and electricity systems, and block chain. Presenting cutting-edge methodologies with real-world application problems and their solutions, the book is useful for researchers, managers, executives, students, academicians, practicing scientists, and decision makers from all around the globe. It offers the academic and the applied communities a compendium and a research resource with significant insights and inspiration for innovative scientific education, investigation and collaboration, to overcome “hard problems” among the emerging challenges today and in the future."

"This volume offers comprehensive coverage of state-of-the-art ensemble learning techniques, including the random forest skeleton tracking algorithm in the Xbox Kinect sensor, which bypasses the need for game controllers. At once a solid theoretical study and a practical guide, the volume is a windfall for researchers and practitioners alike. "

""Deep learning is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts. Because the computer gathers knowledge from experience, there is no need for a human computer operator to formally specify all the knowledge that the computer needs. The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones; a graph of these hierarchies would be many layers deep. This book introduces a broad range of topics in deep learning. The text offers mathematical and conceptual background, covering relevant concepts in linear algebra, probability theory and information theory, numerical computation, and machine learning. It describes deep learning techniques used by practitioners in industry, including deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology; and it surveys such applications as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and video games. Finally, the book offers research perspectives, covering such theoretical topics as linear factor models, autoencoders, representation learning, structured probabilistic models, Monte Carlo methods, the partition function, approximate inference, and deep generative models. Deep Learning can be used by undergraduate or graduate students planning careers in either industry or research, and by software engineers who want to begin using deep learning in their products or platforms. A website offers supplementary material for both readers and instructors"--Page 4 of cover."