• LATEST
    • A PHP Error was encountered

      Severity: Notice

      Message: Undefined variable: latest

      Filename: views/_header.php

      Line Number: 48

      A PHP Error was encountered

      Severity: Warning

      Message: Invalid argument supplied for foreach()

      Filename: views/_header.php

      Line Number: 48


Jaringan Baru AOGNets Untuk Pengenalan Visual Komputer

Sejumlah peneliti pada North Carolina State University (NC State) di Amerika Serikat (AS) telah menciptakan satu kerangka-kerja baru  untuk pembuatan jaringan-kerja yang lengkap, mendalam, dan solid (deep neural networks (DNNs) melalui generator-generator jaringan-kerja yang terpandu tata-bahasa (grammar). Hasil uji-coba jaringan-kerja AOGNets menunjukkan keunggulan dari jaringan-kerja lainnya seperti sistem ResNet dan DenseNet untuk pengenalan visual (Science Daily, 21/5/2019).

“AOGNets have better prediction accuracy than any of the networks we've compared it to. AOGNets are also more interpretable, meaning users can see how the system reaches its conclusions,” ungkap Tianfu Wu, PhD, asisten profesor bidang rekayasa listrik dan komputer pada NC State dan penulis koresponden karya ilmiah AOGNets: Compositional Grammatical Architectures for Deep Learning (Juni, 2019) (North Carolina State University, 21/5/2019).

Karya ilmiah itu dipresentasikan pada IEEE Computer Vision and Pattern Recognition Conference, 16-20 Juni 2019 di Long Beach, California, Amerika Serikat. Penulis utamanya adalah Xilai Li, mahasiswa doktoral pada NC State dan peneliti independen Hi Song sebagai co-author. Karya ilmiah itu didanai oleh The U.S. Army Research Office (hibah W911NF1810295 dan W911NF1810209) (Science Daily, 21/5/2019).

Arsitektur sistem AOGNets menggunakan tata bahasa komposisi berdasarkan praktik terbaik dari sistem jaringan-kerja sebelumnya. Sehingga AOGNets lebih efektif mengekstrak informasi yang berguna dari data mentah (raw data). “We found that hierarchical and compositional grammar gave us a simple, elegant way to unify the approaches taken by previous system architectures, and to our best knowledge, it is the first work that makes use of grammar for network generation,” papar Wu (North Carolina State University, 21/5/2019).

Tim peneliti menguji AOGNets terhadap tiga tolok-ukur klasifikasi gambar: CIFAR-10, CIFAR-100 dan ImageNet-1K. “AOGNets obtained significantly better performance than all of the state-of-the-art networks under fair comparisons, including ResNets, DenseNets, ResNeXts and DualPathNets. AOGNets also obtained the best model interpretability score using the network dissection metric in ImageNet. AOGNets further show great potential in adversarial defense and platform-agnostic deployment (mobile vs cloud),” ungkap Wu (North Carolina State University, 21/5/2019).

Untuk mengetahui kinerja deteksi obyek dan segmentasi semantik contoh, tim peneliti menguji AOGNets dengan tolok-ukur Microsoft COCO yang menggunakan sistem vanilla Mask R-CNN. “AOGNets obtained better results than the ResNet and ResNeXt backbones with smaller model sizes and similar or slightly better inference time. The results show the effectiveness of AOGNets learning better features in object detection and segmentation tasks,” papar Wu (North Carolina State University, 21/5/2019).

Tes-tes tersebut di atas relevan karena klasifikasi gambar termasuk satu tugas pokok pengenalan visual, dan ImageNet adalah tolok-ukur klasifikasi skala besar standar. Demikian pula, deteksi dan segmentasi objek adalah dua tugas pokok visual tingkat tinggi, dan MS-COCO termasuk tolok-ukur paling sering digunakan oleh para ahli selama ini (Science Daily, 21/5/2019).

“To evaluate new network architectures for deep learning in visual recognition, they are the golden testbeds. AOGNets are developed under a principled grammar framework and obtain significant improvement in both ImageNet and MS-COCO, thus showing potentially broad and deep impacts for representation learning in numerous practical applications. We're excited about the grammar-guided AOGNet framework, and are exploring its performance in other deep learning applications, such as deep natural language understanding, deep generative learning and deep reinforcement learning,” ungkap Wu (North Carolina State University, 21/5/2019). 

 

 

Oleh: Servas Pandur